Engaging the public: English local government organisations’ social media communications during the COVID‐19 pandemic

Communication has played a critical role during the initial response to the COVID-19 pandemic, and communicators have had a particularly difficult task in persuading different types of audience to comply with ever-changing regulations. Local government organisations play a crucial role in recontextualising the national messaging for a local audience and encouraging the public to comply with regulations. This paper investigates local government organisations’ (henceforth LGOs) engagement strategies in COVID-related posts on social media. In collaboration with LGOs in England, we examined their communication strategies on Twitter and Facebook during the second UK national lockdown of the COVID-19 pandemic in November-December 2020. Using methods from corpus-assisted discourse studies, the paper analyses the occurrence and functions of selected interactive engagement markers, in this case personal pronouns, questions and hashtags. We find that such linguistic features function to encourage engagement by (a) helping to foster relatedness through ambiguity; (b) creating autonomy-supporting communication; and (c) making messages ‘stand out’. Based on our corpus analysis, we discuss the initial response of the participating councils to our findings and outline future directions including the integration of multimodal approaches to studying the role of localised social media in national crisis management. We argue for more attention to be paid to the many local communicators who play an invaluable role in encouraging the public to comply with national measures in times of crisis.


Introduction
Communication has played a critical role during the response to and management of the COVID-19 pandemic. Traditional news media and social media channels represent key sites of information about the local, national and global news, guidance and policies. However, the proliferation of public service and health promotional messages across many channels, and the amplification of real as well as 'fake news', has led to what World Health Organisation (WHO) Director-General Tedros Adhanom Ghebreyesus labelled as an 'infodemic', a phenomenon just as dangerous as the virus itself ( WHO, 2020 ). Ghebreyesus appealed to social media companies, news organisations and governments to help counter the spread of misinformation and help "sound the appropriate level of alarm " ( WHO, 2020 ). This call is not surprising; the success of pandemic crisis management efforts relies primarily on concerted public action where members of the public have to comply with guidance and regulations. Therefore, communication becomes of crucial importance in providing reliable information and influencing public behaviour valid arguments remains a crucial factor of persuasion, especially with well-informed and active public audiences ( Haugtvedt et al., 1992 ). On the other hand, there are various contextual and socio-cultural constraints, such as the increasing amount of public scepticism towards media sources (e.g. fake news) and politicians. Such constraints require strategic communicators to consider the potential impact of both sourcerelated and audience-related factors on the public reception of strategic messages (cf. Bui et al., 2021 ;Lovari, 2020 ).
The centrality of audience characteristics in determining the choice of message strategies, as well as in evaluating their appropriateness, has been emphasised in public relations, corporate communication and cognate areas ( Rawlins, 2014 ). Previous research on public campaigns has highlighted the diversity of audience characteristics ( McGuire, 2013 ;Parrott, 1995 ). It is because of this diversity that gaining the audience's engagement is often as (if not more) important as presenting them with compelling reasons to support the advocated standpoint. Achieving an appropriate level of appreciation for the relevance of an issue is a prerequisite for encouraging an audience's critical engagement with the information and arguments that are communicated ( Jacobs, 2016 ).
In order to understand how public messaging achieves this aim, it is important to conduct a close examination of previous public science messages, as public campaigners make substantial use of a range of interpersonal strategies to encourage audience engagement, including metadiscourse markers such as pronouns, non-verbal devices (e.g. images) and, in the context of social media, features such as hashtags and emoji ( Martin and MacDonald, 2020 ). However, the close, microlevel analysis of messages on social media remains an under-investigated area within strategic communication research ( Aggerholm and Thomsen, 2014 ;Werder, 2015 ;Palmieri and Mazzali-Lurati, 2021 ). This paper sets out to examine the language of COVID-related social media posts by local authorities in England, focusing on markers of engagement. The aim of this paper is to report on an initial investigation of how micro-level discourse patterns can work as linguistic indicators of communication strategies intended to minimise negative outcomes for local public health. Therefore, the results constitute the basis for further research aimed at examining these strategies on a larger and broader scale and, ultimately, understanding better the role of localised social media in national crisis management.
The specific context chosen for this study is that of local government organisations (henceforth LGOs) in England. These are the county, district, borough and city councils "responsible for a range of vital services for people and businesses in defined areas " ( LGA, 2022 ).
LGOs constitute an ideal terrain for investigating engagement strategies with complex audiences; indeed, socio-demographic factors, cultural and personality traits, political leaning, personal experience with COVID-19, reading and information comprehension and risk aversion behaviour have all been found to influence people's perception of risk, trust and, consequently, their willingness to comply with government guidance ( Coleman et al., 2020 ). 2 Clearly, navigating these complex audiences is an extremely hard task for any communication team, in particular when asking the public to change their behaviour to the extent necessary to reduce the spread of an airborne virus. The broad aim of this work is to better understand how English LGOs tackled that challenge.
The attention to LGOs is warranted for several reasons. Firstly, they play an important role during national crises. As the WHO (2009) have noted, these organisations act as 'translators' of central government communications, in the sense of localising the national messages and helping local residents to make sense of government communication. 2 In the "Pandemic and its public report ", Coleman and colleagues have found six distinct types of population groups differing in their attitudes, experience and behaviour (p. 5): (1) Individualist risk-takers (12% of the population); (2) Noninformation-seeking sceptics (19% of the population); (3) Information-seeking rule-followers (21% of the population), (4) The complacently confident (19% of the population); (5) Information-seeking critics (16% of the population); (6) The experientially risk-averse (12% of the population).
Due to their knowledge of local dynamics, these organisations are able "to provide services in a way people need (and) likely to have a substantially better outcome than through a top-down restrictive framework " ( House of Commons, 2009 ). Secondly, LGOs seem to enjoy much higher levels of public trust. The period before the pandemic has already been characterised by decreasing levels of public trust, and specifically distrust in national level government and leadership ( Edelmann, 2020 ;Enria et al., 2021 ). This trend was also evident during the COVID-19 pandemic when, for example, people were more likely to trust messages coming from their local council than from the national government ( Coleman et al., 2020 ). Despite the evident importance of local government communications, scholarly attention has thus far prioritised national messaging (e.g. Gherhe ş et al., 2023 ;Lovari, 2020 ;Williams and Wright, 2022 ).

Trust and compliance
The persuasive effect of public communication is extremely complex. Public health campaigns in general have the difficult task of influencing resistant audiences, and, as evidence shows, conventional public health campaigns have limited direct effects on health behaviours, although they may exert "moderate to powerful " influence on thinking ( Atkin, 2012 : 13). However, in the case of the COVID-19 pandemic, having an influence merely on thinking has simply not been sufficient; early in the pandemic, the public's active compliance (i.e. change in behaviour) was predicted to be critical to the success of measures brought in to overcome the crisis ( Finset et al., 2020 ). Topics including the complexity of communication aims (with an articulated focus on gaining public compliance), the importance of trustworthiness and ability to engage with complex audiences have already generated considerable research interest. 3 In the below review of existing scholarship, there emerge two particular lines of research: one that examines trust and behavioural influence of COVID-related public health communication and another that focusses on specific communication strategies. amongst the emerging scholarship on COVID-related public health communication, examination of perceived risk, trust and consequent public behaviour are key themes. In their comprehensive report, Coleman et al.,[33][34][35][36][37][38][39][40][41][42][43][44][45][46][47] report on how different groups within the UK public trusted and responded to official guidance on COVID-19 (see footnote 2). What seemed to have been an influential factor in terms of trust was the source of information: people were more likely to trust information when it came to them from scientific resources such as the National Health Service (NHS), the World Health Organisation (WHO) or healthcare professionals (over 88%), or local organisations (72%) than from the national government (63%). In terms of the content of messages, 70% of respondents thought that there was a conflict between government and scientific advice, and 73% thought that government messages were too vague. A closer look at the different types of public audiences gives a more refined picture. For example, 90% of people classed as 'information-seeking critics' (characterised by an 'entrenched suspicion of official advice') found messages too open to interpretation, compared to only 50% of those labelled as 'information-seeking rule followers'. 4 Apart from drawing attention to the general lack of trust in government advice, this report further highlights the importance of communication strategies that engage with specific audiences (see also Section 2.2).
Academic studies provide an overview of the complex relationship between trust, audience and communication in a range of geo-3 Two projects keep an up-to-date list of emerging scholarship: see https://c19 comms.wp.horizon.ac.uk/references/ and https://pandemicandbeyond.exeter.ac. uk/projects/knowing-the-pandemic-communication-information-andexperience/ 4 For more information about the different types of publics please see Coleman et al. (2020) .
political contexts, such as Singapore ( Wong and Jensen, 2020 ), the UK ( Enria et al., 2021 ;Williams and Wright, 2022 ), Italy ( Lovari, 2020 ), Sweden ( Irwin, 2020 ), China ( Zhang et al., 2020 ), Australia and New Zealand ( Bui et al., 2021 ) and the Netherlands ( van Dijck and Alinejad, 2020 ). Lovari's (2020) study of the Italian Health Ministry's communication focuses on social media; it is especially relevant to the present study because of the parallels in terms of the deep distrust in public institutions, combined with the public's growing demand for information both in Italy and the UK. Lovari (2020) examined how the Italian Health Ministry turned to social media, specifically Facebook, to counter the spread of misinformation. The strategies identified include giving voice to influencers, using hashtags, calling out fake news and explaining measures through data and visuals. Lovari concluded that, in a period of extreme uncertainty, public health organisations' use of social media in a transparent, strategic and proactive manner is fundamental to increasing trust.
Another extreme case of trust in public organisations was studied by Irwin (2020) , who examined public communications and international media coverage of the uncommonly liberal pandemic strategy in Sweden, focusing in part on the perception of high levels of trust. Irwin (2020) found that the policies in Sweden were not so different to those in other countries, but what differed was the language and rhetoric relating to the role of social media in the interpretation and ratification of (mis)information. Another example is van Dijck and Alinejad (2020) who -in the Dutch context -reflected on the role of social media in the health crisis and called for a greater understanding of the dual role of social media in both undermining and enhancing public trust, as well as of the importance of developing distinct communication strategies for different aspects of informing and debating with the public.
The above studies seem to present a unified view about the importance of tailoring communications to the needs of various public audiences to gain their trust. Engagement is key in this process because it enhances confidence in the authorities' ability to manage the situation, as opposed to unresponsive, non-transparent communication that leads to the erosion of trust ( Enria et al., 2021 ).
These findings suggest that, to achieve public compliance, communicators need to balance factual information with actions that address the relationship between communicator and audience, for example by communicating at strategically relevant times through: a) source-related strategies, which aim to emphasise the legitimacy of the information by communicating trustworthiness and confidence in the science behind the advice; audience-related strategies, which aim to encourage autonomy, relatedness and empathy.
For example, McGlaughlin et al. (2023) conducted a survey of the UK public's response to various COVID-related public health messages, finding that messaging perceived to be effective provides "a clear rationale for adhering to measures and a means for the public to take personal responsibility to contribute to managing the virus " (p. 14).
Many of the studies listed above and those mentioned in Section 2.1 are similar in that their engagement with communication strategies remains at a 'macro' level; while they provide a broad overview of the strategies of communicators, they do not draw upon systematic and rigorous analyses of individual linguistic patterns. While there is some acknowledgement of the importance of specific linguistic and discourse strategies ( Finset et al., 2020 ;Habersaat et al., 2020 ;Porat et al., 2020 ;Bui et al., 2021 ), discussion of micro-level linguistic strategies is largely absent from the literature. For example, Lovari (2020) notes that the Italian Health Ministry's messages contained emoticons, infographics, and integrated specific words like falso (false) but does not explore in detail these broad observations (p. 460). The exceptions to this include the studies by Gelmini et al. (2021) , Power & Crosthwaite (2022) and Williams & Wright (2022) . Gelmini et al. (2021) , for example, combined the examination of rhetorical appeals with discourse analytical approaches to explore COVID-related corporate communication in Italy, while Williams & Wright (2022) analysed a corpus of televised briefings from the British government, criticising politicians' strategies for minimising their own responsibility for ending the pandemic and maximising the responsibility of the public.
Overall, however, we can say that a number of the observations about language, such as references to "provocative " ( Stolow et al., 2020 , p. 531) or "simple " language ( Finset et al., 2020 , p. 874), for example, lack linguistic precision, and advice given about language use such as "the discourse of crisis, panic and war ", "gain-frame " or positive language ( Haberstaat et al., 2020 , p. 683) lack the specifics that would help professional communicators to apply the advice in practice. Furthermore, much of the existing research on crisis communications in the context of COVID-19 concerns messaging at a national level (e.g. Power and Crosthwaite, 2022 ;Williams and Wright;2022 ). Therefore, the aims of this paper are (a) to contribute to a growing body of knowledge based on detailed accounts of linguistic practices, drawing on empirical data observation of micro-level linguistic patterns, and (b) to explore COVID-related crisis communications at a local rather than national level, investigating how linguistic patterns of engagement may contribute to the communicative goals of English LGOs. As the above review has shown, engagement with a range of audiences and strategic communication are central to achieving trust and public action; therefore, understanding the factors that influence the perception of and engagement with public health measures is key for developing effective interventions in future global crises (cf. Parrott, 1995 ).

Engagement in discourse
The interactions between writers and readers, and specifically the strategies that writers use to engage audiences, have been studied in a variety of contexts, such as academic writing ( Hyland, 2005a ), corporate discourse ( Hyland, 1998 ) or online consumer review discourse ( Vásquez, 2014 ). In language-orientated scholarship, these strategies are commonly referred to as involvement strategies and include resources that encourage interaction between writers and their audiences, and encompass ways in which writers connect with, express concern for, and direct the attention of, their readers (see e.g. Vásquez, 2014 ). Hyland (2017) refers to these strategies as metadiscourse, a concept that describes the language that writers use to help readers interpret the intended function of the message. Metadiscourse links a text to its context by using language designed for readers' needs, understandings, existing knowledge and prior experiences with texts. In applying this concept to the pandemic context, metadiscourse strategies can be said to be clearly very important in pandemic-related health messaging (as discussed in Section 2.2), serving as a "recipient design filter " ( Hyland, 2017 , p. 17) that allow messages to be tailored to specific audiences in order to foster relatedness and encourage autonomy.
In Hyland's model, there are two broad categories of metadiscourse: interactive elements, whose main function is to guide the reader's attention through the text, and interactional elements that aim to involve the audience in the text ( Hyland, 2005b ). For the present study, we have chosen to study a selection of interactional features that foster engagement between writers and the audience and feature in short texts typical of social media. While we acknowledge that a wide range of discursive devices may also serve as engagement markers, our study specifically focuses on: b) • personal pronouns, which are considered as markers of linguistic strategies for engaging multiple voices and communicating trustworthiness (e.g. Aggerholm and Thomsen, 2014 ; Palmieri and Mazzali-Lurati, 2021 ) • questions, which have been shown to function to engage readers through dialogue and may directly influence judgement and behaviour (e.g. Lai and Farbrot, 2014 ;Moore et al., 2012 ); • hashtags, which have been studied as linguistic instruments for engaging readers in discussion of public and societal relevance ( Greco, 2023 ) and shown to take on interpersonal functions as markers of engagement (e.g. Lovari, 2020 ;Zappavigna, 2018 ).
Although these features do not represent the full range of known metadiscourse strategies, our study aims to focus on these features as a window through which to observe some of the patterns of use and communicative functions of engagement strategies in pandemic-related communications by local government organisations.

Data collection
Social media posts from five English LGOs were collected and examined for this preliminary study. The LGOs were: Blackburn with Darwen Borough Council ; 5 Bournemouth, Christchurch and Poole Council ; 6 Oldham Council ; 7 Stockport Metropolitan Borough Council ; 8 and the Royal Borough of Windsor and Maidenhead. 9 In collaboration with the communication teams of these LGOs, all posts from the Facebook and Twitter accounts of these organisations for the period 5 November -2 December 2020 (inclusive) were gathered, representing the period in which England was in its second national lockdown of the COVID-19 pandemic. 10 The selection of participating organisations was necessarily opportunistic; a call was put out through a local government communications consultant, and those authorities that responded positively in time for inclusion in the study were accepted. These organisations differ both in size and social demographic, but this can be considered an advantage, as it provides an (albeit small-scale) insight into a variety of English constituencies; Blackburn, Oldham and Stockport are located in the north of England, while Bournemouth and Windsor are located in the south.
The data were provided by the communication teams of the LGOs, who consented on behalf of their authorities to the use of their posts for research purposes. Both the Facebook and Twitter posts were posted on public channels and are openly accessible to the public. Posts created by private individuals were not gathered; consequently, privacy and ethics concerns related to private individuals did not apply ( Ahmed et al., 2017 ). The collection of the data from the social media sites was completed manually, in some instances by the councils' communications teams, or otherwise by the research team, foregoing any issues related to automatic scraping ( Williams et al., 2017 ).
LGOs who gathered their own data manually were instructed to provide every post published within the specified period, so as to match the collection procedure of the research team, which was to gather all posts and subsequently eliminate those that were not related to COVID-19.
When preparing the data for analysis, the criterion for inclusion of individual posts in the study was the presence of one or more explicit (or clearly implied) references to COVID-19. Explicit reference to COVID-19 was observed through direct mention of the terms COVID-19, COVID and Coronavirus . Implicit reference was judged qualitatively on the basis of posts which contained indirect contextual cues, but did not explicitly mention COVID-19, such as virus, pandemic, social distancing, government guidelines, lockdown, uncertain times and difficult times . Posts that did not refer to COVID-19 (either explicitly or implicitly) were excluded from the dataset. In taking this approach, we acknowledge that reference alone to COVID-19 (whether explicit or implicit) does not guarantee that the topic (or 'aboutness', Scott and Tribble, 2006 ) of a text is centred on the referenced concept, as it is possible that a single reference to COVID-19, for instance, may occur in texts that are ostensibly 'about' a different topic. However, due to the relatively short length of the texts in this study (compared to other commonly analysed texts in corpusassisted discourse studies, such as news articles, for example), it was deemed that even one reference to COVID-19 within a short social media post would very likely indicate that the post is in some way relevant to the topic in question.
Only posts that originated from the councils' social media accounts were gathered, thus excluding posts from other accounts that may have been 'shared' (on Facebook) or 'retweeted' (on Twitter) by the council accounts. In doing so, it is important to acknowledge that some discursive strategies, as evidenced in the sharing of posts from other accounts (see e.g. McEnery et al., 2015 ), may be omitted from the analysis; however, in order to comment on the councils' own engagement strategies, it was necessary to isolate the linguistic content that was authored by council staff. The number and type of posts, as well as the scale of engagement, are summarised in Table 1. 11

Analytical approach
The analysis was conducted using methods from corpus-assisted discourse studies (CADS), an approach that, broadly speaking, combines the quantitative elements of corpus linguistics with the qualitative elements of discourse analysis. The use of corpus linguistics in discourse studies (i.e. CADS) allows access to repeating discourse patterns via the extraction of frequency-based data, which is then analysed and interpreted qualitatively by the researcher; this approach is discussed by Baker (2006) , for example, who provides a general introduction, and Taylor and Marchi (2018) , who provide a critical review of recent developments in CADS. Such approaches have been applied to social media data (e.g. Rüdiger and Dayter, 2020 ;Zappavigna, 2012 ) and have been used to examine language in the context of COVID-related public communications (e.g. Williams and Wright, 2022 ).
The corpus analysis was conducted using AntConc (version 3.5.9; Anthony, 2020 ), a freeware corpus analysis toolkit that is well-suited to handling small datasets such as the corpus investigated in this study. AntConc was used firstly to search for strings that correspond with the 11 We observed differences in the ratio of pandemic and non-pandemic related posts. For example, of the 47 Twitter posts published by Bournemouth, Christchurch and Poole Council, 37 were COVID-related (79%), while Blackburn with Darwen Borough Council published 28 COVID-related tweets out of the 97 during the examined period (29%); however, all 82 tweets posted by Stockport Metropolitan Borough Council during this period related to COVID-19 (100%).  a Questions were accessed through the retrieval of all question marks in the data, rather than searching for interrogative structures. Therefore, we acknowledge that questions that do not make use of question marks are omitted from our analysis. Question marks were searched as escaped characters using regular expressions. b Hashtags were searched as escaped characters using regular expressions. engagement markers included in our study (discussed in Section 2.3); these search terms are listed in Table 2 .
Then, the relative frequency of these terms was normalised to a basis of 10,000. Rather than using a basis of one million, which is common in corpus linguistics, a basis of 10,000 avoids artificially inflating the frequency of features relative to the size of the corpus in this study ( Brezina, 2018 : 43). Relative frequency was used to inform the selection of individual terms for further, qualitative exploration, with a preference for the most commonly occurring terms. This step involved the extraction of all concordance lines of a given term as occurring in both subcorpora, and the manual examination and categorisation of the examples in a spreadsheet. The task of qualitative coding was shared equally amongst the three co-authors with regular review and discussion of each other's' coding decisions. The specific categorisation schemes employed vary according to the terms in question; these are discussed in Section 4.
Following the analysis, we then presented our findings to representatives from the participating local government organisations and held a focus group to discuss their response and feed forward to the next phase of the project. While it is likely that a close observation of data, limited to five research partners, may not yield widely generalisable results, our approach in this study is, nonetheless, to provide an initial overview of some of the discursive engagement strategies in the context of English LGOs. Importantly, the identification of these strategies and their patterns of occurrence may provide a source for the further exploration of possible causative relationships between message and action (for further argument see Grieve, 2021 ). Following the discussion of our initial findings in Section 4, and the response of the participating organisations in Section 5, we discuss -in Section 6 -how this study will inform the next phase of our work.

Findings and discussion
This section presents the analysis of the engagement markers under investigation, starting with overall frequency data (Section 4.1) and then describing the qualitative analysis of some of the most frequently occurring terms for each engagement marker type: personal pronouns and possessive determiners (Sections 4.2.1-4.2.3), questions (Section 4.2.4) and hashtags (Section 4.2.5). Table 3 shows the frequency of each of the search terms in the Facebook and Twitter sub-corpora. For each engagement marker type, the broad distribution of frequency is similar for both Facebook and Twitter sub-corpora, with the exception of the hashtag, which is notably more frequent for the Twitter data; this is expected, as the modern usage of the hashtag as a linguistic meta-tag originated on -and is most closely associated with -Twitter, but has spread to other social media sites including Facebook ( Zappavigna, 2018 ). We refer to the frequency data in Table 3 throughout our analysis in Section 4.2.

Corpus-assisted discourse analysis
The analysis begins by analysing three major categories of personal reference: second person, first person plural and first person singular.

Second person pronouns and agentivity
The second person pronoun you is by far the most frequentlyoccurring personal pronoun in the corpus (see Table 3 ). The importance of you in creating engagement has been well documented in health communication ( Chang, 2011 ;Parrott 1995 ) and social media advertising ( Lai and Farbrot, 2014 ). In analysing the use of you , we examined the following features: • the clause type ( declarative, exclamative, imperative, interrogative and conditional dependant clause ); • the framing operated by the clause ( action/event/situation and subject/object ); • the overarching communicative aim of the whole posted message -in public health communication scholarship and practice there is a distinction between persuasive and informative communicative goals ( Atkin and Rice, 2013 ). Following the close reading and analysis of a sample of our data, we inductively specified further goals within the persuasive category to capture the strength of the deontic modality of the message: advise, encourage, inform, instruct, order .
As shown in Table 4 , The great majority of clauses containing you are declaratives, followed by conditional dependant clauses, imperatives, interrogatives and exclamatives. As for the communicative aim of the posted message ( Table 5 ), ordering prevails, followed by instructing, encouraging, informing and advising.
Furthermore, out of 318 posts aimed at directing and giving orders (across both sub-corpora), only 15 are expressed with imperative   clauses, while the majority (224) are declaratives, followed by conditionals (75). Therefore, the public authorities seem to favour a communicative style that is at the same time official/formal and nonpaternalistic This means favouring strategies that can be described in terms of negative politeness ( Brown and Levinson, 1987 ); compared to imperatives, which may be perceived as explicitly imposing on the reader's freedom of choice, declaratives and conditionals leave the reader space for individual decision-making and minimise interference with "the addressee's freedom of action " ( Brown and Levinson, 2006 : 317). This contributes to the framing of local authorities as reliable and expert sources of information and directions rather than merely promotional agents.
The prevailing use of declaratives may compromise audience engagement compared to using other types of clauses, such as imperatives and exclamatives. Yet, the use of conditionals and interrogatives, which together represent 29% of instances of you across both sub-corpora, indicates that an attempt to engage the readers is present. Indeed, conditional clauses are useful for selecting specific audience groups, attracting their attention and creating a sense of involvement. For example, in Extract 1 shows, readers are invited to verify whether they belong to the category at issue and, if so, to follow the advocated order (compare with observations in Section 4.2.4).

If you are told to self-isolate you must go home immediately ! Only leave your home to go for a test and do not stop self-isolating until you have been given the all clear. This is to stop the spread of #coronavirus
(Oldham Council, 28 November 2020, Twitter) Furthermore, this demonstrates a strategy of synthetic personalisation that is compatible with the mass-mediated nature of the social media post; the use of you simulates a personalised messaging style that encourages engagement by giving an "impression of treating each of the people 'handled' en masse as an individual " ( Fairclough, 2001 : 52)..
As for interrogatives, the questions asked by the writers appear to be intended to (a) stir the curiosity or attention of the audience, inviting them to engage with the content of the post, and (b) personalise the message to the expectation of specific individual readers. Interestingly, the most frequent communicative aim of the posts containing interrogatives is to encourage action (discussed further in Section 4.2.4).
Clauses containing you put the active role of the readers in the foreground, assigning them responsibility. amongst the declaratives, actions in which you is the subject dominate. A frequent pattern (50.36% of all 421 you -subject declaratives in the corpus) is the use of modal auxiliaries with deontic function (e.g. you should, you must ), almost exclusively when referring to actions mandated by COVID-related rules. Of these, the majority (79.25%) order or permit affirmative action (e.g. "Despite national restrictions, you can still exercise outdoors "), while examples of explicit prohibition via negation (e.g. "You must not meet socially indoors with family or friends ") are relatively rare (20.75%). This can be interpreted as an autonomy-facilitating strategy whereby LGOs attempt to remind readers of the freedoms that remain, despite the restrictions in place. There is a low frequency of advising compared to encouragements, where the former entails responsibility on the writer while the latter shifts responsibility on the reader.
Taken as a whole, the posts containing you appear to be structured in such a way to make the readers responsible and interested (cf. Chang 2011 ) in the post while allowing the writers to maintain an image consistent with the ethos of a public authority.

First person plural pronouns and inclusivity
In both the Facebook and Twitter sub-corpora, we is the second most frequently used personal pronoun, behind you . In political discourse, we has been identified as one of the most widely used discursive resources to perform inclusion (e.g. Jaworska and Sogomonian, 2019 ). This is unsurprising -we is a notoriously ambiguous pronoun; it has meanings that can be categorised broadly as exclusive or inclusive . Exclusive usage refers only to the writer (and the people they represent, e.g. "We'd love to hear your stories "). Inclusive usage is, in our analysis, categorised into three types: general (referring to all people, e.g. "We must follow the latest national restrictions "), local (referring to the writer and the reader(s) only, e.g. "These shops will only survive if we continue to spend locally ") (see Darics and Koller, 2019 ), and pseudo 2nd person . Pseudo 2nd person usage occurs when we refers to the reader(s) only and not the writerthis usage does not refer to the writer but instead implies a command (cf. Lammers, 2001 quoted in VAN DE MIEROOP, 2005 ). We coded instances as pseudo 2nd person when it was clear from elsewhere in the tweet that the command is actually addressed to the audience (e.g. "We all have a responsibility to stop the spread of Coronavirus. Remember to: Wash your hands regularly "). The exclusive/inclusive distinction can also be applied to us and our . The ambiguity of the pronominal referent may be used as a strategy to share responsibility for managing the pandemic with the public (cf. Williams and Wright, 2022 ) and mitigate the directness of the command.
In the corpus data, inclusive usage of we comprises the majority of instances of we in both the Facebook and Twitter sub-corpora (58% and 65%, respectively). The pronoun us occurs much less frequently (29 times and 11 times in the Facebook and Twitter sub-corpora, respectively); amongst these instances, inclusive usage of us accounts for 11 instances (38%) on Facebook and 6 instances (55%) on Twitter. The third type of 1st person plural reference we investigated, possessive determiner our , is used inclusively more on Facebook (52%) but relatively less on Twitter (46%).
The predominance of the inclusive usage of we and, to a lesser extent, us and our , clearly demonstrates the effort from councils to create a sense of inclusivity, even in instances where the pronoun refers primarily to the audience and not the writer (pseudo 2nd person) in the guise of a command or order. The ambiguity of the pronoun is an advantage here, used to amplify the sense of inclusion and shared experiences: on the one hand, messages may include several 1st person pronouns with different referents; on the other hand, some pronoun usage is ambiguous by itself, as exemplified in Extract 2. In the first sentence, we and us are used with an exclusive referent, denoting the council; this meaning is further strengthened by the contrast of referring to the constituents ( residents and they ). The referent of we then becomes inclusive ( "please we all need to work together "), reinforced by adverbs and pronouns referring to collective action ( together, each other ). Yet, the use of please in this sentence signals a request or advice, which may indicate that the writer uses 1st person pronouns to mean the audience, not themselves. Subsequently, there are two more shifts in levels of inclusivity revealed by a closer look at the context of the pronouns: after adverb together , the sentence shifts to an exclusive reference to the council ( we believe ) and then again to the shared action of the public ( we can do this ). This may be indicative of an attempt to provide social justification for the council's encouragement of behaviour amongst the public that is ultimately reliant on individual responsibility.
Exclusive references of we , which account for 42% of instances of we in the Facebook sub-corpus and 35% in the Twitter sub-corpus, occur in contexts where councils explicitly refer to their own activities. Exclusive use of we also occurs when the council expresses sympathy or understanding through a personification of the organisation (e.g. we know, we hope , or we believe, please share with us ; see Extract 2) thus projecting the image of a trustworthy, benevolent group of people, as opposed to an abstract organisation ( Fuoli, 2018 ;Palmieri and Musi, 2020 ).
Reflecting on the predominance of inclusive as opposed to exclusive we in the data, the discursive creation of common responsibility can be viewed as a strategy to address "sociable rule-follower " audiences ( Coleman et al., 2020 , p. 14). The sense of inclusivity and shared sense of experiences created through linguistic strategies can also serve to address the public's crisis response, mitigating the emotional extremes, especially outrage ( Malecki et al., 2021 ). Although, at surface level, the exclusive use of the 1st person plural pronoun may not be seen as a strategy to create engagement, the data suggests that, through personification, it helps to discursively create features with the apparent intention to increase trust and consequently encourage compliance.

First person singular pronouns and the hypothetical reader
Even though the 1st person singular pronouns I and me are much less prevalent than those discussed above (see Table 3 ), our analysis reveals a communicative strategy that can be labelled 'hypothetical reader'. Across both the Facebook and Twitter sub-corpora, 54% of the instances of I are used in reference to a speaker who has been created by the communication team itself, often in a mock Q&A format, as demonstrated by Extract 3.

Extract 3 Q: My Favourite pastime is going to the gym. How am I supposed to stay fit and healthy during national lockdown. [sic]
A: The gym might be closed but you can still take unlimited exercise outdoors with your household […] (Stockport Council, 27 November 2020, Twitter) This communication strategy seems to respond to the informational needs of the audience by creating an illusion of bottom-up communication and the co-creation of knowledge. However, since the questions asked in the social media posts have been written by local government organisations themselves, they do not necessarily represent the actual informational needs of their audiences, but rather the 'design' of what these audiences may (need to) be interested in; the questions, presented as part of dialogic interactions between the public and LGOs, may be interpreted as originating from the pro-active listening to people's concerns. This linguistic strategy is autonomy-fostering in two ways: firstly, it creates competence and behavioural change through the internalization of communication ( Porat et al., 2020 ); and, secondly, it solicits the public to take personal responsibility through the construction of the voice of the reader.

Questions and reader engagement
Questions are a highly effective device to achieve communication goals in pandemic-related health communication; they engage readers through dialogue and may directly influence judgement and behaviour ( Moore et al., 2012 ). For written texts, questions are typically rhetorical; they create a semblance of dialogic interaction, without the reader being able to actually respond to the writer ( Curry, 2021 ). On social media, this situation is slightly different, because readers do have the opportunity to respond, although in our dataset the type of questions and their linguistic context (for example that questions are often followed by an answer) seem to suggest that they were not necessarily meant to elicit actual responses. Whether written with a genuine request for information or rhetorically, questions allow authors to share "some of the processes of meaning-making with their readers […], [thus positioning readers] as active participants in the discourse " ( Vásquez, 2014 , p. 107). Previous research has shown that on social media -Twitter particularly -questions (as opposed to statements) lead to a significant increase in engagement with the readership, especially if the questions contain 1st and 2nd person pronouns ( Lai and Farbrot, 2014 ).
Syntactically, questions can be grammatically complete or elliptical, meaning that they contain reduced clauses or phrases ( Carter and Mc-Carthy, 2006 ). Across both sub-corpora, 40% of questions that make use of the question mark are elliptical, typically lacking the modal auxiliary or subject. This usage can be explained by the need for brevitythis is evidenced by the difference between the frequency of elliptical questions on Facebook (where there are no constraints on length) and Twitter (where there is a character constraint) -31% and 60%, respectively. Another (or perhaps parallel) explanation is that digital discourse often mimics spoken language, which is typically more fragmented than writing ( Carter and McCarthy, 1995 ). Such spoken-ness in digital writing has previously been found to create a sense of shared experiences and lead to greater engagement ( Darics, 2020 ).
As Vásquez (2014 : 107) observes in the context of online consumer reviews, questions can serve many functions, including requesting information, expressing suggestions and bringing a topic into focus. The following extracts exemplify the functions identified amongst the 72 questions in our data.
Extracts 4a-4e Extract 4a "It's just a cough, I'll be fine!" Sound familiar? You might think it is 'just a cough' but it could be #coronavirus. (Blackburn with Darwen Council, 7 November 2020, Facebook) Extracts 4a-c show questions that function to draw focus to a specific topic in order to provide information. This is the most common question function, accounting for 83.3% of examples. This strategy appears to be most useful when the information being introduced does not apply to all potential readers but specific subsections. Questions of this type function similarly to conditionals (Section 4.2.1) as focussing devices that appeal to the reader to determine, based upon the criteria encoded in the question, whether they are a member of the targeted subsection, and thus whether the information provided subsequently applies to them. This strategy is used to facilitate reader engagement in the communication of informational propositions, functioning variously to provide advice (34.7%; e.g. 4a), offer support (25%; e.g. 4b) and issue orders (23.6%; e.g. 4c).
Extract 4d is, like 4a-c, a closed question, which acts as a filter of the relevance to the reader of the information that follows. However, unlike 4a-c, the next line is an instruction to provide the information requested by the question, meaning that this is an example of a genuine request for information from the reader, rather than a provision of information by the writer. Requests account for 13.9% of examples.
Extract 4e contains an open question that, like Extracts 4a-c, acts as a preamble to a proposition, in this case information about Remembrance Day celebrations. However, what is notable about 4e is the use of this question to encode a presupposition, defined pragmatically as a proposition that is assumed by the writer to be accepted by the reader (see e.g. Stalnaker, 1974 ). In this example, the presupposition is that readers should plan to celebrate Remembrance Day in a way that complies with current COVID-related restrictions. This is encoded firstly by how , which assumes that the reader will be celebrating Remembrance Day, and secondly by this year , which assumes that the reader already knows that they should celebrate differently than in previous years. Arguably, this is an example of informative presupposition, whereby the writer deploys a presupposition that may not be shared by the reader ( Lewis, 1979 ), the function of which being to persuade the reader to adopt the presupposed idea ( Sbisà, 1999 ). Questions of this type are coded as implicit suggestions, and account for 4.2% of examples.
Another noteworthy observation regarding questions is the voice that is represented. In most cases (91% on Facebook; 96% on Twitter), the voice represented by the question is that of the relevant council. However, the voice of the remainder of questions is implied -as if the Questions are used as a productive resource for generating engagement. The analysis shows that even though the majority (80.5%) are closed 'yes/no' questions, which appear simply to elicit information from the reader, they actually fulfil a range of roles in pandemic health communication that mostly serve to provide (as opposed to gather) information. When not eliciting information, they serve as attention grabbing devices, a role that has been proven to effectively engage readership ( Lai and Farbrot, 2014 ). This is particularly true for questions we identified as focusing on new topics (Extracts 4a-c) and implicit suggestions (Extract 4e). Prompts and suggestions also serve an important role in affecting judgement and behaviour explicitly by highlighting discrepancies between the audience's knowledge and societal/government expectations ( Moore et al., 2012 ), as shown in Extract 4e. Such attention grabbing can help council messages to be more personal and stand out in the social media information overload.

Hashtags and salient information reinforcement
Hashtags are metadiscourse resources typical of microblogging and other social media platforms. Their original function was to create tags that identify topics of discussion, and indeed researchers made use of these identifiers to explore emerging topics during the pandemic ( Petersen and Gerken, 2021 ). However, apart from their role as tags, hashtags can take on a range of communication functions, from experiential functions such as marking topics to interpersonal functions such as providing evaluative metacommentary ( Zappavigna, 2018 ). Structurewise, hashtags can occur independently (at the beginning or end of the social media post) or embedded in the syntactic structure.
Although the relatively low frequency of hashtags in the data (a result of the small size of our dataset) forces us to be hesitant in our conclusions, independent hashtags constitute 55% and 46% of all hashtags on Facebook and Twitter, respectively. While there are several examples of independent hashtags functioning as topic markers, providing a description of what the post is about (for example #coronavirus, #COVID19 and #Diwali ; see Table 6 ), the most common function of independent hashtags, constituting 48% of all independent hashtags, is to perform orders. Hashtags such as #doyourbit, #StayatHome, #StaySafe , and #StopTheSpread have a clearly identifiable imperative structure, and others, such as #HandsFaceSpace , are abbreviated references to orders. Together, in the broader context of the government pandemic crisis communication efforts, these examples can be understood as standpoints -points of view that are defended or justified by means of argumentation ( van Eemeren and Grootendorst, 2004 ). Specifically, these are prescriptive standpoints -they ask the reader not only "to accept the [writer's] evaluation of a particular situation, but also that a certain course of action needs […] to be undertaken in order to change that situation " ( Wackers et al., 2021 : 71). In other words, they say that the current situation requires action to be taken, but not why, because the reader is expected to infer the rationale from the co-text in the post and/or through presupposition of readers' awareness of the pandemic.
The imperative function is amplified when posts feature several hashtags, such as Fig. 1 , where #StaySafe and #DoYourBit have a clear imperative function, the latter repeated in the attached image and with a marked colour distinction, and the third hashtag taking on a function of topic marker.
In a small number of cases (11% of all independent hashtags), we have identified hashtags that provide what Wikström (2014) calls parenthetical or additional information, such as #greatertogether and #Bet-terTogether . In contrast to the imperative hashtags, these can be interpreted not as expressing a standpoint but as expressing arguments in favour of a standpoint. #BetterTogether , for example, is used by Oldham Council in an announcement of local funding from the Greater Manchester Combined Authority for businesses impacted by COVID-19 (Extract 5). Unlike the imperative hashtags, which have an implied subject (the reader), the subject of the parenthetical hashtags is ambiguous; whom or what is 'better together' is ambiguous, even when taking into account the content of the post. Therefore, the hashtag may support one or more of several possible evaluative standpoints -expressions of judgement about facts ( Wackers et al., 2021 : 70). In Extract 5, #BetterTogether may refer specifically to the authorities having intervened to support the survival of local businesses. It may (alternatively, or in addition) appeal to the broader sense of collective action required by all citizens in order to get through the pandemic (see Section 4.2.2 on inclusivity).
Delivered by @GC_BizFinance, there's no need to be an existing customer or open an account.
For the businesses that make Greater Manchester GREAT! https://bit.ly/2WQqYPg #BetterTogether (Oldham Council, 16 November 2020, Twitter) In such examples, the additional information seems to take on a motivational, emotionally expressive force, resulting in the compassionate communication that Finset et al. (2020) and Malecki et al. (2021) define as crucial for the effective management of the pandemic.
Embedded hashtags make up 45% and 54% of hashtags on Facebook and Twitter respectively. In the majority of cases (84% overall), embedded hashtags function as topic markers, where the # symbol acts as a form of punctuation to signal the tag ( Zappavigna, 2018 ). In other cases, embedded hashtags can take on the communicative function of the clause in which they feature; most typically, this means the incorporation of the imperative tags into the sentence structure (for example: "They mean you must #Stayathome as much as possible "). Here too, the hashtag symbol adds an additional markedness to the directive, while simultaneously referencing the broader discourse of the stay-at-home message of the government.

Response of the participating local government organisations
Following our analysis, we presented our findings to communications professionals from the five local government organisations that participated in our study and held an online focus group to gather their feedback.
The fact that our data collection focussed on the second national lockdown was a strategic decision in the hope that, by November 2020 (some nine months into the national pandemic response in the UK), councils would have had time to develop guidelines for COVID-related communication. In reality, only one of our partner organisations had developed such a document. Because of the sudden onset and unprecedented development of the situation, and the extremely high stakes regarding public health, local council organisations were eager to gain some insight about the effectiveness of their, and others', practices.
In response to our findings, the communications professionals were receptive to the opportunity to reflect on their practice and pause to consider how they responded to the challenges of the pandemic response. Especially valuable was the opportunity to compare their communications to those of other councils from elsewhere in England who were dealing with the same challenge but in varying geographical and sociopolitical circumstances. One participant saw value in being shown "the way we can use language and in particular 'you' and 'we' to engage with the audience ", while another appreciated "understanding more about what we do and the science behind it ". A third was excited to "share with the team around use of language, empathy and other key points to help improve what we do ".
Another takeaway from our participants was the sense that, as communications professionals, they had felt largely overlooked and undervalued during the pandemic response, often receiving decisions about national restrictions at the same time as the general public with no advance warning. This, as they reported to us, created a situation where much of the local communication was hurriedly scrambled to keep up with the national messaging. One participant noted that, as a result, much of their COVID-related communication was produced "intuitive-ly…at speed "; therefore, being shown by researchers how linguistic patterns in the data can be ascribed to specific communicative functions made them realise that their work "actually is hugely skilful and valuable ". Related to this is the fact that, while these people were working for local councils to help the public respond appropriately to the pandemic, they were also affected by COVID-19 as personally and emotionally as everyone else and they were adjusting to the ever-changing national restrictions at the same time as the people they were supporting. One participant noted "it's a reminder really that comms doesn't [just] affect our audiences; it affects us too. So staying in the mindset of this for future campaigns is really important. " Overall, the response to our initial study was encouraging. All participants shared an enthusiasm and appreciation for the analysis we conducted and expressed interest in contributing data towards a larger study. We are currently working with these and other local government organisations to gather more data from other key time periods (e.g. the first and third national UK lockdowns, and the 2021-22 wave of the Omicron variant) to explore how communications strategies developed across the first two years of the pandemic in the UK. We have also begun to expand upon our analysis by considering the important role of visual modes of communication in LGO social media posts ( Darics and Love, 2023 ). We have noted, for instance, the presence of a large number of emoji embedded within the texts of the posts, as well as the use of a variety of images that accompany many of the posts. In future, multimodal corpus analysis (e.g. Oakey et al., 2022 ) will be necessary to properly take these communicative tools into account alongside the textual mode. Ultimately, our aim is to reveal to the communications profes-sionals the underlying logic behind their communicative strategies and make our findings available to representatives from LGOs across the UK.

Conclusion
This paper began by outlining the complex nature of public health communication during the pandemic, especially from the point of view of strategic communication, the main aim of which is to achieve public compliance. It has been shown that communicators had a particularly hard task in navigating the 'infodemic' and attending to different types of audiences ( Coleman et al., 2020 ) and communication aims, and this paper set out to provide an initial snapshot into how a small sample of local councils in England were able to navigate these challenges. The analysis was based on the premise that audiences are more likely to comply with the health messaging if they are 'involved' with the message ( Parrot, 1995 ). To this end, the study examined how micro-level linguistic features were used to encourage engagement while helping to address the various publics and fostering relatedness, fostering autonomy and cutting through the 'infodemic'.
One feature that occurred repeatedly in the analysis above is the ambiguity of linguistic resources. This is somewhat counter to the advice previously given in public health communication about avoiding ambiguity ( Parrott, 1995 ), though it has been observed by scholars in pandemic-related communication elsewhere ( Gelmini et al., 2021 ). As we have shown in Section 1, the complexity of pandemic crisis communication is in part the result of the presence of a wide range of audiences, all of whom should be addressed and engaged. The ambiguous and widening referent base of we and us , for example, allows for differing interpretations by the audiences, depending on whether they prefer to be affiliated with the authors of the posts or not.
Ambiguity was also observed in some types of hashtags. amongst the independent hashtags, directive hashtags (e.g. #StaySafe ) appear to serve as explicit commands (cf. Pérez-Hernández, 2018 ), representing prescriptive argumentative standpoints ( Eemeren and Grootendorst, 2004 ;Wackers et al., 2021 ). However, albeit less frequently, parenthetical hashtags ( Wikström, 2014 ) such as #bettertogether demonstrate the interpersonal, evaluative functions of hashtags ( Zappavigna, 2018 ). They put forward arguments in support of ambiguous standpoints, thus relying on readers to supply their individual interpretations. However, hashtags used to perform the most common function in our data, topic marking (e.g. #coronavirus ), are unambiguous; they explicitly reinforce connotations related to the overall message, making key information salient, which reflects the LGOs' efforts to cut through the 'infodemic', while appealing to a range of audiences.
There is also evidence of how the councils used autonomy-supporting communication strategies, which, according to Porat et al. (2020) , lead to autonomous motivation, and are more likely to lead to sustainable change. One such strategy is the use of questions which, although not often inviting actual responses, nonetheless involve readers in the meaning-making process by giving them the sense of interactive, readerinvolved engagement (cf. Curry, 2021 ). The analysis of 2nd person pronoun you has also shown similar efforts, whereby deontic posts were predominantly articulated in the form of encouragement, thus shifting the responsibility to the readers. Both in conditional sentences (Section 4.2.1) and in questions encouraging desired behaviours through presupposition (Section 4.2.4), readers were encouraged to individually interpret their experience and verify for themselves whether it matched the scenario hypothesized in the post and take responsibility for the consequent behaviour. The use of 1st person pronouns in what was identified as mock Q&A provided a voice for the reader (albeit a hypothetical one), creating a sense of personal responsibility and encouraging the internalization of the messages.
Finally, the analysis shows the councils' efforts to balance an image consistent with the ethos of a public authority with strategies that make information and guidance stand out in the 'infodemic'. Messages used a range of attention-grabbing devices (questions and mock Q&A), visual markedness (hashtags) and discourse strategies to appeal to a shared sense of physical experiences (spoken features). Through use of direct address ( you ) and inclusivity ( we ), a prevalent number of social media posts used synthetic personalisation to encourage readers to interpret the guidance as having direct relevance to them -this process has previously been found to successfully facilitate active engagement and compliance with public health messaging ( Parrott, 1995 ).
Perhaps the most important realisation is that the examined metadiscourse devices -first and second person pronouns, questions, and hashtags -take on a range of concurrent interactive functions that make official social media crisis communication trustworthy, interesting, relevant and relatable: the four exact message quality features Atkin (2012) calls for in persuasive health communication. Although the analysis in this paper has only been able to provide a snapshot of select linguistic features, it nonetheless provides scholars and practitioners with an insight into the importance of exploring micro-level language phenomena in strategic communication. We hope that the linguistic and discourse strategies shown in this paper may serve as concrete examples that provide a basis for reflection for communication practitioners so that they can craft messages with a greater chance of success in mobilising the public.
The response of both our communications consultant and representatives from the councils who participated in this study reinforce the crucial role that local organisations had in mediating and 'translating' messaging from government communications. The communication teams of these local government organisations found themselves under immense pressure. While personally battling through the unprecedented times of a global pandemic, they had to respond professionally in an unfamiliar communication context, working with oftentimes problematic, ambiguous government messaging (e.g. Williams and Wright, 2022 ). As we learned from our participants, their work very much relied on instinctive responses to national regulations. As our work with these and other local government organisations continues, we aim to develop communicative guidelines to help these previously under-appreciated communications professionals feel better supported in advising their local public in times of crisis.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.