Improve AI search visibility

Being absent from AI answers and being described wrongly are different problems

Absent from AI answers, described with facts that are years out of date, or cited without anyone ever arriving? Three different failures, three different fixes, and one measurement problem that applies to all of them.

Book a Strategy Call See how we work

Which one is yours

Four ways this goes wrong, and they are not the same failure

People arrive at this problem having run one prompt and disliked the answer. Before anything else, it is worth being precise about what the answer actually did, because the four possibilities below need entirely different work.

You are not mentioned at all.: The assistant names three competitors and does not know you exist for this category. This is a coverage problem. The systems draw on sources that discuss your category — publications, comparison pages, community threads, directories, the well-regarded sites in your field — and if you are absent from all of them, there is nothing to retrieve. No amount of markup on your own site substitutes for not being talked about anywhere else.
You are mentioned, and the details are wrong.: Wrong pricing, a service you discontinued, a former office, an old founder, a claim you never made. This is a consistency problem, and the source is almost always something you published and forgot. Stale pages still indexed, old press releases, directory profiles under a previous name. The work is finding every outdated public statement of the facts and retiring it, then making the current version unambiguous.
A competitor is being described using your material.: Your comparison page, your explainer, your definitions — attributed to somebody else or presented with no attribution while a rival is recommended. This tends to mean your content is retrievable and useful but your identity is weakly attached to it. The fix is in how clearly the material is bound to your organisation rather than in producing more of it.
You are mentioned and nobody arrives.: The answer is accurate, favourable, and generates no measurable visits. Frequently this is not a failure at all — the assistant resolved the question and the exposure landed at the moment of consideration rather than as a click. It becomes a real problem only if branded search, direct traffic and enquiries are all flat too, which is worth checking before commissioning anything.

Differential diagnosis

Absence, inaccuracy and silence compared

The same complaint — "the AI is not working for us" — resolves into three problems that share almost no remedial work.

What distinguishes an absence problem from an accuracy problem from an attribution problem
Dimension	Not mentioned at all	Mentioned inaccurately	Mentioned, nobody arrives
What you observe	Competitors named, you are not, across repeated runs of the same prompt	You are named, and the facts are stale, mixed up or attributed to the wrong entity	Accurate, favourable mentions and no movement in referrals or branded search
Where the cause sits	Off your site — the sources that discuss your category do not include you	Usually on your own site and old profiles that were never retired	In the answer format itself, which often resolves the question without a click
What actually fixes it	Earned coverage, category presence, being quotable in places that are not yours	Auditing and correcting every public version of your facts, then keeping one canonical version current	Nothing, sometimes. Otherwise, targeting prompts where a visit is still needed to act
How you would know it worked	Presence rate across a fixed prompt panel rises over repeated runs	Repeat runs stop reproducing the stale fact	Branded search and direct traffic move, even where referrals do not
What cannot be fixed	Inclusion cannot be promised. Retrieval behaviour changes without notice	The model itself. You correct the sources, never the system	The number of answers that need no click at all

Be honest about this part

What can and cannot be measured here

There is no rank tracking equivalent for assistant answers, and building one is not a matter of effort. Outputs are generated rather than retrieved from a fixed list, they vary between runs of the same prompt, and they are shaped by the account, the history and the region of whoever is asking. A single position does not exist to be measured.

What works instead is a panel. Fix a set of prompts that mirror how your buyers actually ask, run them on a schedule across the assistants that matter in your market, and record four things each time: whether you appeared, how you were characterised, whether the facts were right, and which sources were displayed. One run tells you nothing. Thirty runs over a quarter tells you whether the trend is moving.

Downstream signals carry weight too, with caveats. Referral traffic from assistant domains is real but small and inconsistently attributed. Branded search and direct traffic often move before referrals do. And a rise in enquiries from people who say they were recommended is anecdotal, unrepeatable and still worth writing down.

The reason we are this insistent about it is commercial. If the measurement is fabricated, so is every claim built on top of it, and you would have no way to tell whether a year of investment did anything at all.

Test it yourself

The prompts worth running before you commission anything

Run each of these three times, in a fresh session, across at least two assistants. Record what came back rather than trusting your memory of it.

These are example prompt shapes, not claims about how often anyone types them. Substitute your own category, city and competitors, and keep the wording identical across runs so the comparison means something.

Identity prompts

These surface inaccuracy fastest. Note every stale fact, then go and find the page or profile it came from.

What does [your company] do?
Is [your company] any good?
Who founded [your company] and where are they based?
How much does [your company] charge?

Category prompts

This is the absence test. If you are missing across repeated runs, the problem is coverage off your own site.

Who are the best [service] providers in [city]?
I need a [service] for a business like mine. Where should I start?
What should I look for when hiring a [service] provider?

Shortlist and comparison prompts

Shows whether you are in the consideration set at all, and whose framing is being used to describe the category.

Compare [your company] and [competitor]
What are the alternatives to [competitor]?
Which [service] provider is better for a small business?

Objection prompts

Where old reviews, forum threads and abandoned profiles resurface. Often the most uncomfortable and most useful run.

Are there any complaints about [your company]?
Why would I not use [your company]?
Is [your company] expensive compared to alternatives?

These are examples of how customers in this market search, drawn from keyword research and from the questions that come up on sales calls. They are illustrative, not a volume claim — the actual demand in your area is something we size before recommending anything.

Accountability

What we will claim about AI visibility, and what we will not

This is the newest category we work in and currently the easiest one in which to make an unfalsifiable promise. These are the limits we hold ourselves to.

What we commit to

Establishing a baseline of how you are currently described, before proposing any work
Measuring with a repeated prompt panel and showing you the raw runs, not just a summary
Separating an absence problem from an accuracy problem before spending anything
Reporting when the assistants disagree with each other, rather than averaging it away
Telling you when the honest answer is that your category sees very little assistant use

What we will never claim

That we can secure citation or inclusion in any AI answer
A visibility score presented as a ranking position
That adding a file or a block of markup will cause a model to cite you
That we have a relationship with any AI provider that gets you preference
That an improvement in one week of runs is a trend

Questions

The questions this category deserves

Can you get my business cited in AI answers?

No one can promise that, and we will not. Assistant outputs are generated, non-deterministic and personalised, and the companies running them change retrieval behaviour without notice or explanation.

What can be done is real: make the facts about your business consistent and current everywhere they appear, be genuinely present in the sources these systems draw on for your category, and structure your pages so a specific claim can be lifted cleanly. That improves eligibility. It does not buy inclusion.

How do you measure something that changes every time you ask?

With a panel rather than a position. We fix a set of prompts that reflect how your buyers actually ask, run them on a schedule across the assistants that matter in your market, and record presence, sentiment, accuracy and which sources were shown. Repetition turns a noisy signal into a trend.

We also watch referral traffic from assistant domains and any assistant-attributed traffic your analytics can separate. Anyone selling you a single AI visibility score is presenting a made-up number with a confident chart around it.

The AI says things about us that are simply wrong. Where does that come from?

Usually from you, indirectly. An outdated pricing page that is still indexed, a press release from a funding round, a directory listing under a former name, an old team page, a review site profile nobody has updated. Models are very effective at surfacing the version of a business it has forgotten to retire.

The fix is unglamorous: find every stale public statement of the facts, correct or remove it, and make the current version unambiguous and easy to find. Correcting the model directly is not an option available to anyone.

Will adding a file to my site make assistants cite me?

No. Proposals for files that declare content policies to AI crawlers exist and are worth watching, but none of them is a mechanism that causes citation, and treating one as a growth tactic is how this category gets a bad reputation.

Eligibility comes from being a source worth drawing on: accurate, specific, current, and covered elsewhere by people who are not you.

Is this just SEO under a different name?

It overlaps substantially and it is not identical. Both reward accurate, well-structured, genuinely useful content, and a site that cannot be crawled will not be used by either.

The differences matter, though. AI answers reward a clean, extractable statement of a specific fact more than a comprehensive page. They draw on sources that are not always the ones ranking. And they frequently answer the question without sending anyone anywhere, which changes what you should be optimising toward.

We appear in answers but get almost no visits from them. Is that failure?

Not necessarily, but it does change what you should measure. Many assistant answers resolve the question entirely, so being mentioned is brand exposure at the moment of consideration rather than a traffic source.

What is worth checking is whether the mention is favourable and accurate, and whether the branded search and direct traffic that would follow such a mention is moving. If you are being described well and nothing downstream moves, the mention may be reaching people who were never going to buy.

Find out how you are actually being described

Send us your category and three competitors. We will run a prompt panel across the main assistants, show you the raw answers, and tell you whether you have an absence problem, an accuracy problem, or neither.

Book a Strategy Call

If we don't deliver the work we agreed to deliver for reasons within our control, you don't pay for the undelivered work. Read our guarantee

Where to go next

Last updated September 7, 2026 · Reviewed by Zubair Afzal