How to Know If Your Organisation's Public Narrative Is Accurate in an AI-Driven World

A black and white target with a red dart hitting the bullseye.

Organisations could assess their public narrative before AI by reading the press coverage and listening to customers. If the gap between the two grew wide enough, someone in communications noticed and adjusted the messaging. 

People now meet an organisation without visiting its website. They ask an AI system a question, read the summary, ask a follow-up question, then compare the answer with a rival's. An impression forms before anyone reaches your homepage, and no one in your building approved a word of it.

So how do you know if the version of your organisation now circulating is accurate?

One approach communications teams take is to ask if AI systems are being fair to them. I think accuracy is the more useful test, and it has two halves: does the public record support what you claim, and are your claims true?

Your official narrative is only one part of the story

Your intended narrative is often in the website copy, social media, annual report and whatever the press office has been briefing. It says what you do and how you differ from competitors. However, the public meets your brand via customer reviews, employee posts, old news coverage, regulatory findings and comparisons written by competitors who had every reason to be unkind. LLMs draw on all of it and produce a single confident paragraph.

You can be completely accurate in your own communications and still be badly represented outside them. A company might have delivered a major programme, but if no independent source has written about the work, an LLM has no reason to treat the achievement as part of its identity. Your corporate narrative works better as a claim waiting to be tested than as a finished product.

What I would argue for

• Start by separating facts from interpretation

Corporate language mixes different kinds of statements without flagging the difference. "We operate in 15 markets" can be checked. "We are a market leader" depends entirely on how you define leadership. Take your main public messages and ask what evidence would convince a sceptic. If the answer comes down to personal belief, you’re probably describing ambition that can’t be verified.

• Ask AI what it thinks before telling it what you want it to think

If you feed a model your own website and ask for a summary, you have tested little beyond its ability to read and summarise your content. Ask open questions without giving it your own description first, and you get a better sense of the story the wider information environment produces. Ask what you are known for and what criticisms are associated with you. Ask what it is like to work there and then read the answers as a stranger would.

• Trace the answer back to evidence

An AI answer is only as useful as the information behind it. If a model calls you innovative, don't take the compliment at face value. Find out why it reached that conclusion. It might be drawing on a recent award, or it might be repeating an article from 2018 that has been cited so often that it now looks like established fact. The same applies to criticism. If several systems repeat the same negative claim, that doesn't necessarily mean several independent sources reached the same conclusion. They may all be drawing on the same original story. Before you decide that you have an AI problem, find out where the claim came from, so you don’t risk fixing the symptom instead of the cause.

• Accuracy is not the same as completeness

An answer can contain zero errors at all and still leave a false impression. For example, a company may have made real progress on emissions but still be strongly associated with a past pollution incident. If an AI summary keeps returning to the incident and barely registers the progress, every fact is correct, but the overall picture is still wrong. The reverse can happen too. A flattering reputation can bury limitations that customers know about perfectly well. That is why hunting only for factual inacurracies is only half the challenge

• Build a scorecard that measures what’s important

Rather than giving the organisation one overall score, assess the strength of each of your six to ten most important public claims and rate them from 1 to 5. This gives you a clearer view of where your narrative is well supported and where it needs more work:

• Factual accuracy: Is the claim correct and backed by current evidence?

• Source quality: Is it pulled from independent or owned sources?

• Completeness: Does the public record include what changes how the claim should be read?

• Proportionality: Is the weight given to achievements and criticisms fair?

• Consistency: Do different AI models reach similar conclusions?

• Internal alignment: Does your own employee and customer evidence support the claim?

Treat AI narrative monitoring as an early-warning system

Public narratives rarely change overnight, but they can shift gradually through small signs such as new criticism in industry conversations, competitors becoming better known for something you were once known for, or an old perception of your organisation continuing after the business has changed. AI can help spot these changes by bringing together information from different sources and showing the story as it evolves.

One bad answer does not tell you much, so look for patterns across different AI systems and over time. If several systems consistently describe your organisation in the same way, the signal is stronger, while wide variations suggest that the public narrative is fragmented.

That is why I would use AI monitoring as an ongoing exercise and monitor trends. The most useful finding may not be an obvious error but an early sign that people are starting to see your organisation differently, perhaps months before traditional reputation research picks it up.

Which brings me back to the question I opened with. You'll know your public narrative is accurate when it holds up without you there to explain it. When independent sources broadly support what you say, and your own people recognise the organisation being described, the story can stand on its own. AI has no view on whether that story is true, but it will make it much easier for people to see where the story you tell and the organisation they experience don’t quite match.