Who is this for? What are we actually saying? Can we back it up? And who will notice when it stops being true?
I spent a day at BrightonSEO listening to talks about AI search, and kept coming back to those four questions.
The Brighton Centre was packed. There was an otter in bright pink, a seagull wearing a conference badge and a lot of people excited about where search is heading. Best believe I have photos with both mascots on my phone reel. It felt good to be among peers, with time to think about the work beyond whatever needed approving that afternoon.
Between the demos and the debates, something kept surfacing: getting content found depends on decisions about what it says, how it’s organised and whether anyone is keeping it accurate.
The technology gives us plenty to investigate. Some of the questions it exposes are already sitting in our content backlog.
Being found is only part of the job
Search has always involved losing some control over how your content appears. Headlines get shortened. Snippets lift a sentence out of a page. A journalist quotes part of a report. Nobody owes your carefully constructed introduction a reading.
AI-mediated search now adds another stage of interpretation. A system can retrieve information from several sources, combine it and offer an answer that becomes someone's first encounter with your organisation. Google describes how its AI search features may draw on several related searches and supporting pages.
That matters whether you publish product information, research or press statements. Someone might be asking which service suits their business. They might be checking what your organisation said about an issue, or looking for evidence to support a decision.
Your content needs to be discoverable, and the information someone encounters needs to make sense beyond the page where you arranged it.
One distinction worth keeping from the day was between being cited, being mentioned and being recommended. A system might cite your guide to explain a subject without recommending your business. It might mention your business while explaining why another option would be more suitable.
Those are different outcomes. Counting them together can make a dashboard look reassuring while leaving the important question unanswered: what is someone learning about you?
In Auditorium 1, Jean Sarunporn of MG OMD offered a thought experiment:
If your website disappeared from the internet, would there still be enough evidence for AI to recommend your brand?
That question stayed with me. It asks what supports your claims beyond your own copy: independent reporting, reviews, other people's experiences. A very confident About page can’t do all that work by itself.
The writing problem
Later, in Auditorium 2, Alex Moss of Yoast described reading three sentences of an About page and still not knowing what the company did. His version was rather less polite.
The problem was immediately recognisable. Imagine a fictional software company whose website promises a “flexible, market-leading platform for growing businesses”.
It sounds plausible. It’s probably survived several rounds of approval. I still don't know whether it will do what I need.
Can I change the number of licences each month? Does it work with the system I already use? Is support included? What happens if I leave?
“Flexible” might be entirely justified. The writer's job is to explain how.
That takes access to people who know the product, the patience to ask follow-up questions and the judgement to recognise when an answer is incomplete. Replacing “flexible” with a more energetic adjective will not get us very far.
Paris Childress’s talk on information gain gave me another way into this: what knowledge does a piece of content add? As an editorial question, that earns its place in a brief.
What can we tell someone that they won't find in the next five articles on the same subject?
It could be original research. It could be an expert explaining why a familiar approach fails under particular conditions. It could be a worked example, or a clear answer to a question customers keep asking because nobody has explained it properly.
Much of that material may already exist inside the organisation. Finding it is part of the work. So is checking it. In my work, asking where a stat came from is a reflex. A number doesn’t become evidence because it’s made it into a nicely designed slide. Customer conversations can reveal valuable questions, but using their contents requires appropriate permission and care.
AI can help organise source material and produce a draft. The brief still needs to say which evidence is approved, which qualifications matter and what remains uncertain.
Good writing makes those distinctions understandable. It gives the reader enough substance to decide whether something applies to them, including when the answer is no.
The agreement problem
Now imagine our software company gets more specific.
The product page says customers can cancel at any time. The FAQ says customers are committed for a year, with no early cancellation. A downloadable guide describes a monthly plan that the company stopped selling last year.
Each item may have been accurate when it was published. Together, they make a rather unhelpful answer to a straightforward question. Everyone can defend their own page. The reader has somehow become responsible for resolving the disagreement.
Which terms would a prospective customer encounter? Which version might appear in search? What would an AI-generated comparison make of the discrepancy?
Before anyone starts optimising, somebody needs to establish which terms apply to which plan, for which customers, and from what date.
The structured data sessions brought this into focus for me. The question underneath the technical detail was about relationships: how do the things a business describes connect to one another?
A company provides a service. That service has particular conditions. A report contains findings from a defined study. A person wrote the report on behalf of an organisation. Making those relationships explicit helps us describe what we actually mean.
Schema markup is one way of expressing information in a machine-readable form. It needs to represent the content accurately. Google’s guidance says structured data should match the visible page content, and that no special schema is needed for its AI search features.
Adding markup to contradictory information doesn't settle the contradiction. The organisation still has to agree on the facts.
Agreement doesn’t mean every page must use identical wording. A press statement and a technical guide serve different readers. They should be able to explain the same subject differently without disagreeing about what happened, what is available or what the evidence supports.
This is where content strategy becomes very practical. Work out which source is authoritative, what needs to be connected and where a qualification belongs. Then make those decisions visible in the content.
The machine should not be the first participant asked to reconcile the business's position.
The ownership problem
Resolving the contradiction is a start. Keeping it resolved requires someone to own the answer.
In our fictional company, a change to subscription terms needs to reach the product page, FAQ and downloadable guide. It may also affect sales materials and support content. Updating whichever page someone happens to remember is an unreliable maintenance plan.
A workable process identifies who confirms the change, who assesses its consequences and who updates the affected content. It also makes clear what happens to material that should no longer be used.
A quarterly review date can help. A change to the underlying service should trigger a review when it happens.
The automation sessions offered practical ways to take some of the labour out of this: checking for changes, comparing versions, flagging missing information and bringing scattered data together. These are tasks worth making easier.
People still need to decide whether the revised explanation is accurate and sufficient. A check can tell you that a field contains text. Someone needs to determine whether the text answers the question.
For writers, this is consequential work. The qualification you carefully negotiated into a sentence needs to survive the next edit, the next format and the next product change.
Measurement has three stages, and they’re easy to blur: the update went live; the information appears accurately where people look for it; something improved for readers or the business. More citations tell us content is being referenced, not that it is being represented accurately; someone still has to read the answers. Fewer repeat questions to support and clearer enquiries get closer to business impact. The next question is what caused the improvement.
Decide what improvement means before the chart starts going up.
Start with one answer
You don't need to begin by rebuilding the entire content estate.
Choose one important question someone asks about your organisation. Find the places where you answer it. Follow the links, open the PDF and check the FAQ that hasn't attracted anyone's attention for a while.
Do the answers agree? Is the evidence sound? Are the conditions clear? Who needs to act when something changes?
That gives you somewhere to start, with a specific answer to improve and a process to keep it accurate.
I came back from BrightonSEO with things to investigate. I also came back with a familiar question worth asking a little earlier.
Before we make this easier to find, have we agreed on what we're saying?