Discovery in search and AI answers
How your brand, your business and the people in it are found and described in search and AI answers, and what to do about it. We work with marketing teams, brand and reputation teams and the analysts who research companies.
The world has changed with AI
People's behaviour has changed. They ask AI at any moment, anywhere, for almost anything, and they ask it the questions they once typed into a search box. The answer they get names a few brands, describes them in a sentence or two and leaves the rest out. Discovery now means being in that answer as much as being on the results page.
Traditional search and the basics of SEO still matter. Google still sends most of the traffic, and the pages that rank well are the pages the AI models read. But the work now has to serve what AI needs equally, and that is different. AI needs plain, well-structured information that fits any database and any index, published at a scale that matches how a model learns, because a model does not read your best page once. It samples everything it can find and repeats what it finds most often.
What you may be experiencing
Every engagement starts with what you are seeing, because the symptom tells us where to look. These are the themes we come across most, grouped by the three things a brand needs in search and AI answers. Each group speaks to a different team, and most clients recognise themselves in more than one.
Are you in the answer at all?
- Traffic or revenue from search has fallen and nobody can say where it went.
- New customers, readers or audiences are not reaching you, and you do not know how to cut through to the ones that matter.
- A rival is named for the questions you should own.
Does the answer describe you correctly?
- AI answers state incorrect facts about your products, your prices or your history.
- The description is inconsistent. One model says one thing and another says the opposite.
- You are described for something you no longer do, or not for the thing you do best.
Is the answer fair to you?
- A past problem is repeated to everyone who asks, long after it was fixed.
- Your category has a reputation challenge and the models pin it on you.
- You need to know what AI says about the company and its leaders before an investor, a journalist or a buyer asks.
How a discovery project works
We take each theme in turn and address it with a plan, and we use AI end to end to do it. That is our point of difference. Most agencies use AI for the content and keep the research and the measurement by hand, which is slow and expensive and cannot keep up with answers that change every week.
- We use AI to build the research. We ask the models the questions your customers ask, at each step of their journey, and we measure where your brand, your rivals and your people stand in the answers and on the results page. We read the pages the answers cite and audit your content against the competitor's. The counting is done outside the AI, and the tools are AI and platform agnostic, so the research reads any model and any search engine the same way.
- We use AI to build the solution. From the research we write the plan: what to publish, where it needs to be picked up, which incorrect facts to correct at their source and in what order. We draft the structured data and the content outlines with AI, and your team or agency delivers them.
- We embed AI in your processes. We leave a measurement running on the schedule you choose and a routine the team owns, so the change lasts, runs efficiently and works at the scale the models need.
This draws on twenty years of search, media and data work for global brands, and on our own research into what AI models say about companies.
The user's journey, made concrete
The journey framework is only useful when it is tied to real questions. The example is a UK energy supplier, a category where both households and businesses buy. The list gives the question people ask at each step and what the AI answers did with one brand. The table beside it holds the figures, and the chart beneath compares whose pages the answers cite at each step, which is where the content audit against a competitor starts.
- Learn"What should I look for in an energy supplier?" The answers describe the category, and the brand's own guides are never cited.
- Shortlist"Which three suppliers should I consider for a small business?" Usually second or third, and for price rather than service.
- Verify"Is this supplier any good for a heat pump tariff?" One model in four states a tariff that is a year out of date.
- Execute"How do I switch to this supplier?" The answers cite a comparison site, not the brand's own switching page.
- Remember"Should I stay with my supplier or switch?" The objection raised most is a past complaints ruling.
| Step | Named | Incorrect fact | Own page cited |
|---|---|---|---|
| Learn | 22% | 4% | 2% |
| Shortlist | 54% | 9% | 6% |
| Verify | 38% | 25% | 9% |
| Execute | 71% | 11% | 4% |
| Remember | 45% | 14% | 7% |
Questions people ask us
Is SEO dead now that people ask AI?
No. Most traffic still comes from search, and the pages that rank are the pages the models read. What has changed is that ranking is no longer enough. The same page has to be clear, structured and cited widely enough that a model repeats it. SEO is now one workstream among several that have to be run together.
What are AEO and GEO, and do we need a new agency for them?
AEO, answer engine optimisation, is being the answer when a search engine or an AI answers a question directly. GEO, generative engine optimisation, is being named and described correctly by models such as ChatGPT. They are new names for work that overlaps heavily with good search and content work. You do not need a new agency. You need one plan that covers all three, and a measure that shows whether it is working.
Can a company control what ChatGPT says about it?
Not directly, and anyone who says they can is overselling. What you can do is correct the facts at the sources the models cite, publish the plain, structured information the models learn from, and measure on a schedule whether the answer moved. Most incorrect facts come from pages the company does not own, so fixing your own site is necessary but not enough.
Is discovery in AI answers different for a B2B business?
The journey is longer and the questions are more specific, so a B2B brand is often cited in an answer without being named. The research has to find those cases and work out who is named instead and why. Everything else is the same. The models learn the same way whoever the buyer is.
Why does structured data matter so much for AI answers?
Because a model samples at scale and repeats what it finds most consistently. Information that is laid out the same way on every page, fit for any index, is what survives that sampling. Prose on one well-written page does not.
If you would like a reading of where your brand stands, our details are on the contact page.