object(WP_Post)#9099 (24) { ["ID"]=> int(2179) ["post_author"]=> string(2) "36" ["post_date"]=> string(19) "2019-06-25 01:55:18" ["post_date_gmt"]=> string(19) "2019-06-25 01:55:18" ["post_content"]=> string(9657) "

A headline might be a reader's first – and only – contact with a brand, and many will keep skimming until they land on something that takes their interest.

If you aren't into the nitty-gritty of headlines, stop reading now. But if you want to be that content creator who writes the runaway headline, here's a snapshot of what some of the research has found.

Between 1 March and 10 May 2017, BuzzSumo analysed 100 million of the most shared article headlines on Facebook and Twitter, the platforms dominated by publisher and consumer content. Then in July, it published its analysis of 10 million B2B headlines – those shared on LinkedIn – and found that the best headline phrases, structures, numbers and lengths differed from the B2C results.

1. What works for B2C content

While previous research suggested that the first three and last three words were the important parts of a headline, the BuzzSumo research highlighted linking phrases as key for headlines targeting B2C audiences.

The three-word phrase – or trigram – that led the engagement charge (likes, shares, comments) was 'will make you'. In fact, on Facebook it had twice as many engagements as the trigram that took second place ('this is why'), followed by 'can we guess', 'only X in' and 'the reason is'.

BuzzSumo determined that the success of the 'will make you' phrase was based on it linking content to the emotional impact it will have on the reader – it sets you up to care ('will make you cry', 'will make you smarter', etc.).

It also found that headlines that provoke curiosity work well when readers are looking to learn something from an article. They are a little like the 'will make you' articles, but they tell you what you'll find out rather than what you'll feel.

The BuzzSumo research found that the top five phrases starting a B2C headline were:

1. X reasons why…
2. X things you…
3. This is what…
4. This is the…
5. This is how…

The top five phrases ending a B2C headline were:

1. …the world
2. …X years
3. …goes viral
4. …to know
5. …X days

Admittedly, the second-place holder might not rate as well in Australia, but the five top-performing first words were:

1. This…
2. Trump…
3. How…
4. 10…
5. Why…

So, what doesn't work for B2C audiences? The five worst-performing frequently used phrases were:

1. control of your
2. your own business
3. work for you
4. the introduction of
5. what's new in

Confirming earlier Outbrain research, BuzzSumo found that 12 to 18 words and 80 to 95 characters had the highest engagement on Facebook.

2. What works for B2B content

In BuzzSumo's analysis of 10 million headlines of articles shared on LinkedIn, the practical and informative nature of how-to and list posts (see #3 below) proved to be strong performers in the top five most popular three-word phrases:

1. X ways to…
2. The future of…
3. X things you…
4. How to get…
5. How to make…

There was a clear frontrunner in the top two-word phrases starting headlines – 'How to…' was shared almost three times more on average than the second-place holder. The top two-word phrases starting B2B headings were:

1. How to…
2. The X…
3. X things…
4. X ways…
5. Top X…

Note that after the 'How to…' phrase, the next four most shared phrases were all forms of list posts, which gained more than double the average shares of ‘what’ or ‘why’ posts.

Celebrity brand names also garnered high levels of engagement. It makes sense that companies influencing the business environment and forging technological and business model innovation – like Uber, Google, Apple, Facebook, Tesla and Amazon – will have strong reader appeal. For example, nib's Ambulance or Uber: Who you gonna call?generated a lot of conversation on its Facebook page due to Uber's topicality.

At seven to 12 words, the optimum headline length for LinkedIn is much shorter than for Facebook.

3. The ongoing power of the list

In July 2017, CoSchedule founder Garrett Moon published results of an analysis that began with close to one million blog headlines – which were then put through various filters. The top takeaway was that list posts or listicles (headlines that start with a number) are "huge". Moon wrote they are "the most likely type of post to be shared 1000 or even 100 times". Interestingly, he also noted that "list posts only made up 5% of the total posts actually written".

The BuzzSumo research, confirming the power of lists and the list post format, found the six most effective numbers (in descending order) in B2C content are 10, 5, 15, 7, 20 and 6. In B2B content, the most shared numbers that start post headlines are 5, 10, 3, 7, 4 and 6, with 5 and 10 performing equally well. Note that how-to posts outstripped list posts in B2B.

CoSchedule's results show that list posts that they identified by the words 'thing', 'should' and 'reasons' – '5 things you can do…', '4 reasons why you should…' – do best on Facebook, Twitter and LinkedIn.

It's possible that this is due to a combination of clear promise (‘10 steps’, etc.) and the scannable nature of the post, where you can easily work out which bits you want to read.

4. Emotion is good but beware the bait

While strong emotional headlines and those provoking curiosity may get you results, you need to rein in any urge to overstate.

In May 2017, Facebook announced it would demote “headlines that exaggerate the details of a story with sensational language” and those that aim “to make the story seem like a bigger deal than it really is.”

There may be some debate about what is and isn't clickbait, but there are two key points to consider. In the first place, the reader needs to feel encouraged to read. And in the second, they need to not be disappointed when they have finished reading.

5. Research, tailor and test

There are no hard and fast rules. You always need to research what works for your audience, your topics and your social platforms, and to test your headlines. Different audiences will require different content and will be accessing it on different platforms. For example, Outbrain works for an editorial-led audience more than a business-specific audience.

In the interests of transparency, this headline isn't the first that came to mind. It's the result of trawling through this research.

Maybe we all need to take the advice of Ann Handley, chief content officer at MarketingProfs: "Spend as much time writing the headline as you do an entire blog post or social post."

Belinda Henwood, Strategy & Content

" ["post_title"]=> string(35) "5 reasons why a headline goes viral" ["post_excerpt"]=> string(159) "A headline might be a reader's first – and only – contact with a brand, and many will keep skimming until they land on something that takes their interest." ["post_status"]=> string(7) "publish" ["comment_status"]=> string(4) "open" ["ping_status"]=> string(4) "open" ["post_password"]=> string(0) "" ["post_name"]=> string(35) "5-reasons-why-a-headline-goes-viral" ["to_ping"]=> string(0) "" ["pinged"]=> string(0) "" ["post_modified"]=> string(19) "2019-06-26 00:53:19" ["post_modified_gmt"]=> string(19) "2019-06-26 00:53:19" ["post_content_filtered"]=> string(0) "" ["post_parent"]=> int(0) ["guid"]=> string(43) "https://isentiastaging.wpengine.com/?p=1842" ["menu_order"]=> int(0) ["post_type"]=> string(4) "post" ["post_mime_type"]=> string(0) "" ["comment_count"]=> string(1) "0" ["filter"]=> string(3) "raw" }

object(WP_Post)#11906 (24) { ["ID"]=> int(2178) ["post_author"]=> string(2) "36" ["post_date"]=> string(19) "2019-06-25 01:53:05" ["post_date_gmt"]=> string(19) "2019-06-25 01:53:05" ["post_content"]=> string(4889) "

Is content marketing an art or a science?

It’s not a new debate but an increasingly relevant one. As technology continues to improve, the C-Suite is demanding a clearer measurement into impact. Marketing and communications professionals responsible for curating content are no longer governed by ‘gut feeling’ and instead, are increasingly driven by engagement metrics to demonstrate ROI.

These professionals are well aware how their role requires a mix of art and science thinking. They both draw from the left brain and the right brain, using data and reason to guide the creativity that fuels it.

But this relationship is less rigorously applied to content marketing – an emerging discipline that straddles both marketing and communications objectives.

Marketers and communications professionals have varying levels of social media sophistication – particularly with LinkedIn, which is often a core channel for content. With LinkedIn estimating more than 130, 000 posts are made on its newsfeed every week, organisations are increasingly turning to it as a distribution channel for thought leadership.

Far fewer, however, understand how to draw insight from the platform to ensure their content connects with their target audience.

Marketers and communications practitioners will often speak to me with this challenge solely in mind. Most are able to gauge the success of content on Facebook and Instagram to some level. Plenty of tools exist which measure various social aspects of content marketing, such as ‘likes’ or ‘shares’. But real engagement isn’t buzz. Determining whether content is connecting with a target audience is a key challenge.

Content marketers are struggling to understand whether their current LinkedIn strategy is working – whether it’s reaching the right audience and whether a piece of content is being actively engaged on the platform.

Other times, they will want to target a particular demographic; millennials for example. But they don’t have the understanding of what this group is looking for when they log onto this social networking site.

In short, what content marketers want to do is debunk the myths surrounding their own activity and drill down into strategy to make their dollars work harder.

How can data help?

Data is pivotal. Armed with information, marketers and communications professionals have a window into the opinions, passions and motivations of their audience.

At Isentia we’ve seen this in our own business. The Research & Insights stream has grown by 25 per cent in the last year, as this market recognises the importance of data. I’m often told by clients that they’re just at the start of their measurement journey, but still desperately rely on data to convince the C-Suite to spend money on content marketing.

Research & Insights can be used to help inform content marketing strategy by highlighting what brand-relevant topics an organisation’s audience is engaging with. It can also help content marketers understand where their brand sits against those their competitors, by measuring their share of voice on a particular topic.

But most importantly, data can help marketing and communications practitioners build out content itself. By understanding what type of content receives the most engagement on the platform, they can tailor their content strategies and measure their success at the same time.

Data is the key to debunking the myths of what does or doesn’t work in a content marketing strategy. It gives marketers and communications professionals the opportunity to ensure they understand their audience first and foremost, in order to communicate in a way that connects.

This is where science can help inform the art in content marketing.

Asha Oberoi
Head of Insights, Australia 

" ["post_title"]=> string(27) "Is your content connecting?" ["post_excerpt"]=> string(41) "Is content marketing an art or a science?" ["post_status"]=> string(7) "publish" ["comment_status"]=> string(4) "open" ["ping_status"]=> string(4) "open" ["post_password"]=> string(0) "" ["post_name"]=> string(26) "is-your-content-connecting" ["to_ping"]=> string(0) "" ["pinged"]=> string(0) "" ["post_modified"]=> string(19) "2019-06-26 00:54:22" ["post_modified_gmt"]=> string(19) "2019-06-26 00:54:22" ["post_content_filtered"]=> string(0) "" ["post_parent"]=> int(0) ["guid"]=> string(43) "https://isentiastaging.wpengine.com/?p=1841" ["menu_order"]=> int(0) ["post_type"]=> string(4) "post" ["post_mime_type"]=> string(0) "" ["comment_count"]=> string(1) "0" ["filter"]=> string(3) "raw" }

object(WP_Post)#8878 (24) { ["ID"]=> int(935) ["post_author"]=> string(2) "36" ["post_date"]=> string(19) "2019-06-20 00:50:35" ["post_date_gmt"]=> string(19) "2019-06-20 00:50:35" ["post_content"]=> string(610) "

Video content represents 80% of all internet traffic in 2019

Video content continues to rise in popularity. We have explored how marketers can connect with their video audience and drive strong engagements.

Download our whitepaper to learn more.

" ["post_title"]=> string(53) "A marketer’s guide to connect with a video audience" ["post_excerpt"]=> string(97) "Find out the importance of video content in 2019 and how you can connect with your video audience" ["post_status"]=> string(7) "publish" ["comment_status"]=> string(4) "open" ["ping_status"]=> string(4) "open" ["post_password"]=> string(0) "" ["post_name"]=> string(50) "a-marketers-guide-to-connect-with-a-video-audience" ["to_ping"]=> string(0) "" ["pinged"]=> string(0) "" ["post_modified"]=> string(19) "2023-07-07 02:41:43" ["post_modified_gmt"]=> string(19) "2023-07-07 02:41:43" ["post_content_filtered"]=> string(0) "" ["post_parent"]=> int(0) ["guid"]=> string(35) "https://isentia.wpengine.com/?p=935" ["menu_order"]=> int(0) ["post_type"]=> string(4) "post" ["post_mime_type"]=> string(0) "" ["comment_count"]=> string(1) "0" ["filter"]=> string(3) "raw" }

object(WP_Post)#11881 (24) { ["ID"]=> int(48129) ["post_author"]=> string(2) "75" ["post_date"]=> string(19) "2026-06-09 02:26:42" ["post_date_gmt"]=> string(19) "2026-06-09 02:26:42" ["post_content"]=> string(19927) "

Following our webinar on 5 May, our panelists respond to the questions we didn’t get to on the day.

How comms leaders need to adapt to this new AI shift at the workplace?

AI is already shaping your organisation’s reputation — whether you’re managing it or not.

On 5 May, Isentia brought together three leading voices in communications and insights for a conversation about what’s really happening on the ground as AI reshapes the workspace. Catherine Arrow (Executive Director, PR Knowledge Hub), Russ Horell (Isentia APAC’s ex-Chief Revenue Officer) and Ngaire Crawford (Isentia and Vuelio's Executive Director for AI Strategy in PR & Comms) explored how communications leaders are navigating AI conversations with executives and boards, where pressure is increasing across risk, measurement and strategic advisory, how teams are adapting workflows and decision-making in response to AI influence, and where do communicators see the right opportunity.

The session saw many questions popping up from our audiences that we couldn't really address them all. So we went back to our panelists and asked them to respond. Below, Catherine Arrow and Ngaire Crawford share their thoughts on what attendees most wanted to know.

Catherine Arrow, Executive Director, PR Knowledge Hub

Catherine Arrow is the Executive Director of PR Knowledge Hub, a professional development and training organisation for public relations practitioners. A veteran of the communications industry with deep expertise in strategic counsel, crisis and issues management, and information disorder, Catherine is known for her clear-eyed thinking on the intersection of AI, reputation and organisational responsibility. She is a trusted voice on what AI actually means for practitioners — not in theory, but in practice.

Q1. Comms professionals often have an idea of how AI can help us, but often the C-Suite have other (less informed) ideas. Do you have examples of how you’ve tactfully pushed back or diverted focus back to where you feel it should be (outcomes focused)?

One of the main difficulties is that organisations and their leaders seldom have a clear picture of what they already have at their fingertips when it comes to AI. Many organisations, for example, use the Microsoft suite and may already have access to Copilot, but what can actually be achieved depends on the licences, payments and subscriptions in place. At the same time, leadership teams are influenced, as we all are, by the level of hype that has bubbled to the surface over the last 12 months. Too often, AI is regarded as a passive tool that lives inside a box and as practitioners we have a role to play helping leaders move beyond that limited view. We need to help them understand not only the functional use of particular tools but the bigger picture, to understand the impact AI may have on the organisation’s decision-making, relationships, reputation and licence to operate. The issue is whether the organisation understands the consequences of handing decisions, or the appearance of decisions, to AI in ways that may affect stakeholders, employees, communities of interest and others connected to the organisation’s activities.

So, when I need to tactfully push back or redirect the conversation, my starting point is usually a set of simple questions. What are you trying to achieve with this? How does it align with your organisational outcomes? Is it being applied ethically? Do you understand the consequences? What could it do to your reputation, relationships and ability to maintain your licence to operate?

That approach allows the conversation to move away from the excitement of the new shiny tools and back towards purpose, responsibility and organisational impact. From there, you can begin to workshop the options, discuss the implications, consider the real costs and identify the areas that need attention before AI of any kind is deployed.

Q2. How much is AI picking up on social media commentary as part of its description of organisations?

Yes, AI picks up social media commentary but it will only pick up what it can access. Generally, that means publicly available commentary or material available through an API connection or approved data source. So, in terms of general digital chatter, yes, AI can identify and interpret some of that activity.

The difficulty is that we have to be careful about what it is actually reading. You can already see this in some AI overviews and AI-generated summaries, where the system may refer to “chatter” or online discussion without always digging deeply enough into whether the original sources are genuine, reliable or themselves AI-generated. So we end up with AI nested inside AI, nested inside AI.

That creates a bigger problem for communication and engagement. People are increasingly using AI to generate and optimise social media content but that is not the same as engaging with people. At the same time, many platform algorithms are designed to reward optimised content. The result is a circular loop where AI feeds AI, which feeds AI again. Human language, judgement and connection get pushed aside.

People can become immune to this kind of content because it does not sound like the way we speak to each other, nor does it reflect the way genuine relationships are built. Then, when conflict or outrage is layered on top, the environment becomes even harder to interpret.

So the short answer is yes, AI can monitor social media commentary. The longer answer is that it often does so in ways that require considerable caution, human judgement and a much deeper understanding of what is being surfaced, amplified and missed.

Q3. How are you maintaining credibility in a landscape flooded with AI-generated content?

Personally, I try to maintain credibility by doing my best to remain human. That is probably the best advice I would give to others as well. Use your own intelligence to understand the people and communities you want to engage with. Do not use AI as a barrier between you and them. Use it as a handy tool. Let it help you edit where necessary, test an idea or explore an angle, but do not hand over your voice, judgement or identity. The same applies to imagery. If you are creating images with AI, treat it as a collaboration rather than giving the system an idea and simply running with whatever it gives back. AI-generated imagery carries assumptions and bias, so we must question what is produced and make conscious choices about what we use.

For me, maintaining credibility and authenticity means being yourself and not allowing AI to suffocate your identity. That will become harder to do as digital twins, synthetic voices and other tools make it easier for organisations to use it as a mask. The real challenge is not so much maintaining credibility. It is about maintaining humanity, empathy, kindness and a genuine wish to connect with others beyond the AI-intermediated space.

Q4. Globally, it would be interesting to learn how each country’s culture is reflected in the messaging as filtered by LLMs.

Different AI systems can reflect, distort or flatten cultural context in several ways and one of the biggest concerns is the continental drift between the major model providers. Many of the systems most widely used are strongly shaped by US language, culture, law, commercial assumptions and social norms. At the same time, Chinese models are being developed within a very different political, linguistic and cultural environment – much better at APAC languages for example. So the question is twofold: whether an AI system is “accurate” and “accurate according to whom, trained on what, governed by which assumptions and optimised for which worldview”?

Training data matters enormously. In the early days of the general release of generative AI, we saw certain words and phrases appear everywhere. “Delve” is one example, and “dive into” is another. These were signals of the linguistic patterns embedded in the data, the training process and the reinforcement layers shaping outputs. When those patterns are repeated at scale, they begin to influence the way people write, speak and frame ideas. Over time, that blunts understanding, with distinctive voices, local idioms and cultural ways of knowing pushed towards a generic machine-mediated style.

There is important work being done by Māori researchers and others on the cultural impact of AI, particularly in relation to language and data sovereignty, indigenous knowledge and the right of communities to determine how their knowledge is represented, protected and used. The research is still developing but the concern is real. AI systems can absorb, repackage and reproduce cultural knowledge without context, consent or accountability. They can also misread or flatten concepts that do not translate neatly into dominant languages or Western knowledge structures.

That is why the homogenisation of culture and language is something we need to understand and contest. In many ways, AI becomes a form of digital colonisation. Knowledge is scraped, curated, classified and reproduced by systems that may have no meaningful relationship with the people, histories or communities from which that knowledge came. In some instances, it risks rewriting history, or at least a narrowing of it, where contested, local or marginalised perspectives are buried beneath the most available, most optimised or most dominant version of events.

So, different AI systems may distort cultural context by privileging dominant languages, simplifying complex meanings, mistranslating concepts, omitting local histories or reproducing the worldview of their developers and training environments. They may flatten culture by making everything sound the same. And that presents a real danger, not only for communication professionals but for society more broadly, because shared understanding, cultural memory and social cohesion all depend on our ability to recognise difference, preserve nuance and respect the knowledge that communities hold for themselves.

Q5. Where can we find Catherine’s upcoming sessions on misinformation and AI?

The Managing Information Disorder session will stream live on 2nd July. Please register here.

In case you can't make it, you can always signup and access the live recording. As part of the session, you will also receive the Information Disorder Framework and the practical tools that accompany it, designed to help you recognise and respond to misinformation, disinformation, mal-information, narrative attacks, deepfakes and other risks in the current information environment.

If you would like to know more about AI, the AI in Public Relations – What’s New, What’s Next and What Now? session is also available. It is designed to help you get up to speed with the latest developments, understand what they mean for public relations practice and identify what you need to do next.

You can also access some of the resources Catherine mentioned during the webinar, including the Chaos Compendium, which is freely available. It exists to help you think through what is happening now, prepare your organisation for the months ahead and take practical steps to manage the risks, issues and pressures already coming into view.

Ngaire Crawford, Executive Director, AI Strategy

Ngaire Crawford is Executive Director for AI Strategy, with a mandate spanning both Isentia and Vuelio to ensure the Group’s AI strategy is coherent, credible and commercially effective. A driving force behind Isentia’s insights and measurement capability for a number of years, Ngaire is a well-respected voice across the communications measurement industry — with customers, at industry events, and in the broader conversation about the future of PR and communications. Her curious, thoughtful approach, deep expertise in measurement, and early adopter mindset with AI have helped shape much of what Isentia is building.

Q1. What are some of the top errors or mistakes you see communications leaders make in regards to AI?

If we assume people are already off the first rung and past treating AI as a workflow assistant for drafting and summarising, the more interesting mistakes tend to start after that.

The one I’d put first is assuming this is a more neutral information environment than it actually is. It’s a tempting thing to believe after years of algorithmic outrage, the idea that AI hands everyone a calmer, more balanced version of events is genuinely appealing. But I don’t think the echo chamber disappears with LLMs; it just gets dressed differently. Social platforms built echo chambers by amplifying whatever made you angry. LLMs have a gentler version of the same habit, they’re built to be helpful and agreeable, so if you ask a leading question you’ll often get an answer that politely validates your framing. And the more personalised they get, the more pronounced that becomes. So when you’re thinking about how your audiences are forming views through these tools, what matters isn’t just what the system “says” — it’s who’s doing the asking, how they’re asking, and what the system has already learned about them.

And then a more practical one: getting the order of operations wrong when you build out intelligence capability. The instinct to bring more of this in-house is understandable, but it often gets handed straight to a data or tech team, and however good the pipeline they build, you can end up with something impressive that produces information nobody quite knows how to act on. What’s signal versus noise for this organisation, what’s actually useful to a comms leader — are communications questions, not engineering ones. Sort those out first and the technology tends to slot in behind them; do it the other way round and you usually get the impressive-but-unusable version.

Q2. Would it be accurate to say content with an overt evidence base will “perform” better in an AI information environment?

The thing is, “perform” is doing two jobs. There’s visibility (does evidence-rich content get cited more?) and there’s reputation (when you do get cited, is the picture the system paints one you’d actually recognise?) They’re not the same question, and an evidence base does fairly different things for each.

On visibility, it’s, broadly yes. Well-sourced, clearly structured, quotable content does tend to get picked up more, there’s research pointing that way, though honestly it’s mostly from controlled studies and it moves around a lot depending on the topic and the platform. But what’s getting rewarded there is just clarity, good sourcing, consistency, authority. Which is less a shiny new lever and more the basics of communications.

Reputation is where “perform better” can start to lead you astray. Getting cited isn’t the same as being represented well. You can have a flawless evidence base, get pulled into an answer, and still find that answer describes you in a way you’d never have approved because the model’s also leaning on everything everyone else has said about you. You can definitely nudge your visibility, but how you’re represented is downstream of your whole information environment, and that’s a slower, longer term shift.

So yes, a real evidence base matters, but not because it’s a button you press to perform better. It matters because being genuinely worth referencing is what trusted sources cite, and it’s those sources, built up over time, that shape how these systems talk about you. What I’d be wary of is treating an “overt evidence base” as something you manufacture to game your way in.

The conversation continues

What comes through clearly in both Catherine’s and Ngaire’s responses is that AI is a shifting set of conditions that communications professionals need to understand, question and actively work within, not just hand over.

The organisations that will navigate this well are not necessarily those with the most sophisticated AI tools. They are the ones asking better questions earlier, about purpose, about accountability, about what it means to remain genuinely credible and human in an environment where both are increasingly easy to fake.

---

If you missed the webinar or want to revisit it, access the recording here. Watch this space — there’s more to come.

" ["post_title"]=> string(44) "Inside the AI Shift: Your Questions Answered" ["post_excerpt"]=> string(189) "Panelists from Isentia's "Inside the AI Shift" webinar address some of the audiences' unanswered questions on maintaining credibility, AI leadership and evidence-based content performance. " ["post_status"]=> string(7) "publish" ["comment_status"]=> string(4) "open" ["ping_status"]=> string(4) "open" ["post_password"]=> string(0) "" ["post_name"]=> string(43) "inside-the-ai-shift-your-questions-answered" ["to_ping"]=> string(0) "" ["pinged"]=> string(0) "" ["post_modified"]=> string(19) "2026-06-09 02:26:47" ["post_modified_gmt"]=> string(19) "2026-06-09 02:26:47" ["post_content_filtered"]=> string(0) "" ["post_parent"]=> int(0) ["guid"]=> string(32) "https://www.isentia.com/?p=48129" ["menu_order"]=> int(0) ["post_type"]=> string(4) "post" ["post_mime_type"]=> string(0) "" ["comment_count"]=> string(1) "0" ["filter"]=> string(3) "raw" }