October 15, 2026

Most people find this out the same way. They ask ChatGPT who the best supplier of something is in their area, half expecting to see themselves, and get a list of four competitors instead. Two of them are smaller. One of them you have never heard of.

It is a genuinely unpleasant moment, and the conclusion people draw afterwards is usually wrong. They assume it is about quality, or that the system favours big brands, or that it is simply random. It is almost never any of those three.

There are eight common reasons. They are diagnosable, and most of them are fixable in weeks rather than months. This article works through them in the order you should check.

First, Is It Actually True?

Before diagnosing anything, test properly, because one question asked once on one day tells you very little.

Ask the same question in four places: ChatGPT, Perplexity, Gemini and Google AI Overviews. Ask it with web search switched on and switched off, because they retrieve differently and the difference tells you which problem you have. Ask it the way a buyer would, in plain language, without your company name anywhere in the question. Then repeat it across a few days.

Write down who gets named and which sources each answer cites. Three patterns tell you three different things:

• You appear sometimes and not others. You have a consistency problem rather than an absence problem. Your signals exist but they are not strong or uniform enough for a machine to rely on.
• You never appear but your sector does. A visibility problem. Something specific to your business is stopping you being named, and it is almost certainly on the list below.
• The assistant cannot answer the question sensibly at all. The category is early. That is the best position on this page to be in, because being first to fill a quiet space is far cheaper than displacing somebody later.

The Eight Reasons, in the Order to Check Them

1. Nothing independent exists about you

This is the most common cause by a distance. If every piece of information about your business originates from your own website, an assistant has nothing to corroborate against, and these systems are cautious about naming a business purely on its own say-so.

The fix is not more blog posts. It is being mentioned somewhere that is not yours: trade press, industry publications, podcasts, the directories your sector actually respects, partner sites, genuine participation in communities where your buyers spend time. One substantive independent mention moves this further than twenty pages on your own site.

This is also the reason the fix feels unfamiliar. Most businesses have fifteen years of habit pointing at their own website. Authority in AI search is largely built somewhere else.

2. Your details disagree with each other

Your business name on your website, on Companies House, on your Google Business Profile, on LinkedIn, and in directories you were listed in years ago and forgot. If those disagree, even slightly, a machine cannot confidently decide they describe the same organisation, so it reaches for one it can.

Old addresses. A trading name that differs from the registered name. Three different descriptions of what you do. A phone number still live somewhere that nobody has answered since 2019. This is tedious to fix and it moves things quickly, which is an unusually good combination.

3. You are invisible to Bing

ChatGPT retrieves live web results primarily through Bing. Plenty of UK businesses have spent fifteen years optimising for Google and have never once opened Bing Webmaster Tools. If Bing does not have you indexed properly, ChatGPT will struggle to find you no matter how well you perform in Google.

Registering and submitting a sitemap takes about twenty minutes and is the single fastest technical win available to most businesses on this list.

4. Your content cannot be extracted

Assistants quote passages, not pages. If your site buries the answer, writes around the subject, or presents everything as flowing prose with no structure, there is nothing clean to lift. Meanwhile a competitor with a weaker business and a clearer FAQ page gets quoted in your place.

The test is blunt. Open any important page and ask whether a stranger could find the answer to the page’s core question in the first two sentences. If not, that page is unquotable, however well written it is.

5. You say nothing specific

Vague copy is unquotable copy. If your website contains no numbers, no defined method, no timeframes, no named people and no checkable claims, there is nothing for an assistant to state as fact and attribute to you.

Businesses resist this, reasonably, because specifics feel like hostages to fortune. Publishing a timescale means somebody can hold you to it. But the alternative is being unquotable, and unquotable now means invisible. Pick the specifics you are comfortable defending and publish those.

6. Nobody is named

Anonymous content is treated as weaker than attributed content, and that gap has widened considerably. In February 2026 Google added an Authors section to its documentation and a core update days later rewarded named authorship. If your posts are written by “Admin” and your About page lists no actual humans with actual credentials, there is no expert for an assistant to attribute anything to.

We go through this properly in how to get AI to recognise you as an expert, because for professional services in particular it is often the whole problem.

7. Everything is old

Recency influences what gets cited. A website whose most recent content is two years old signals a business that may not be trading, let alone leading its field. This is unfair to a great many excellent businesses, and it is how these systems behave.

You do not need to publish weekly. You do need visible evidence that the business is active and that the information is current.

8. Your server is turning the crawlers away

The one almost nobody checks. Your robots.txt may welcome every crawler while your firewall, security plugin or host quietly rejects requests that do not look like a human browser. AI crawlers send minimal headers, which is precisely the pattern some security rules are configured to block.

Ask whoever manages your hosting to check the server logs for GPTBot, OAI-SearchBot, PerplexityBot and ClaudeBot, and to confirm they are receiving normal responses rather than being refused. We have seen sites inviting crawlers in at one layer and turning them away at another, and nobody noticed for months because nothing looks wrong from the outside.

The Two Reasons People Assume, and Why They Are Usually Wrong

“It is biased towards big brands”

Understandable, and mostly untrue. Large brands do appear often, because they have more independent coverage and clearer entity signals, not because size is a ranking factor. The mechanism rewards verifiability, and large organisations happen to be easier to verify.

This distinction matters enormously. The conclusion you draw from the bias theory is that nothing can be done. The conclusion you draw from the verifiability explanation is that a great deal can be done, relatively cheaply. In narrow categories, small specialists routinely beat national brands in AI answers, for exactly this reason: they are easier to name confidently in response to a specific question.

“It is random”

It varies, which is not the same thing. Answers differ between accounts, between days and between models, because these are probabilistic systems retrieving live information. But across enough checks, patterns emerge and the same names keep appearing.

Test the same question across four assistants over several days before drawing any conclusion at all. What looks like noise at one data point usually resolves into something consistent at twenty.

Which One Is Mine?

A rough diagnostic, based on what you saw when you tested.

What you observed	Most likely cause	Where to start
Named sometimes, not others	Weak or inconsistent corroboration	Reasons one and two
Never named, competitors always are	No independent presence	Reason one, then five
Named in Gemini but never in ChatGPT	Bing indexing	Reason three
Named, but described wrongly	Inconsistent or thin information	Reasons two and five
Nobody in your sector is named	The category is early	Publish first, you have an opening
Strong site, still absent everywhere	Crawler access or extractability	Reasons eight and four
Named for the wrong service	Unclear service structure	Reasons four and five

What Do I Actually Do in the First Fortnight?

Ordered by speed rather than by importance, because early movement makes the slower work much easier to justify to whoever holds the budget.

1. Register with Bing Webmaster Tools and submit your sitemap. Twenty minutes. If ChatGPT is where you want to appear, this is the most disproportionate return available to you.
2. Make your business details agree everywhere. Website, Google Business Profile, LinkedIn, Companies House, every directory you can find yourself in. Same name, same address, same description of what you do.
3. Check the crawlers are getting in. Ask whoever manages your hosting to confirm the AI crawlers are receiving normal responses in the server logs. Five minutes of their time, and occasionally it explains everything.
4. Put names on your content. Author, role, credentials, and Person and Organisation schema behind them.
5. Add FAQ markup to the questions you already answer. Most sites have the content and none of the markup, which means a machine has to infer a structure you could simply hand it.
6. Pick one page and make it genuinely specific. Three vague claims out, three checkable facts in. Do it once, see how it reads, then work through the rest.

Then leave it a month and re-test. Not because a month is long enough for everything, but because you need a second data point before you can tell whether anything you did mattered.

How Do I Know If It Is Working?

This is where most businesses come unstuck, because the reporting habits of the last fifteen years do not transfer. There is no position one to screenshot and no ranking table to send the board.

Four things are worth tracking monthly:

• Named or not named, across the same fixed set of buyer questions, in the same four assistants. Keep the questions identical each month or you are measuring different things and calling it progress.
• How you are described when you are named. Being mentioned inaccurately is a different problem from not being mentioned, and it needs a different fix.
• Which sources the answers cite. When your own pages start appearing in that list, something is working.
• AI referral traffic and what it converts at, tracked separately from the rest of your organic traffic. This is the number that decides whether any of it was worth doing.

Keep a simple record. A spreadsheet holding the date, the question, the assistant, who was named and which sources were cited will tell you more after three months than any tool will tell you on day one.

How Long Does This Take to Fix?

It depends which reason you have, and the range is wide enough to matter.

The technical and consistency items, reasons two, three, four, six and eight, can show movement within two to six weeks. They are not about becoming more authoritative. They are about letting authority you already possess be read properly, which is why they move so quickly.

Reasons one, five and seven, the ones concerning independent presence, specificity and freshness, take six months and beyond, because they involve genuinely changing what exists about you in the world. There is no shortcut, and anyone offering you one is selling something.

Most businesses we work with see the first shifts inside ninety days and meaningful change over six to twelve months. That is the same shape as SEO, for the same underlying reason.

Is It Worth the Effort for a Small Business?

A fair question, and the honest answer depends on how considered your buyers’ decisions are.

If people find you and buy within minutes, largely on price, this matters less. If they research first, compare a few options and want to feel confident before making contact, it matters a great deal, because the research stage is exactly where assistants have inserted themselves.

Most businesses selling anything considered fall into the second group and have not noticed the change, because the enquiries that never arrive are invisible. Nobody emails to tell you they chose a competitor an assistant recommended. You simply see a quiet month and blame the market.

How We Work Through This at TU

We run the diagnosis rather than guessing at it. The same questions across four assistants, with and without live retrieval, and a written record of who gets named and which sources the answers lean on. That source list is usually the most useful thing to come out of the exercise, because it shows you where the conversation about your industry is genuinely happening, and it is rarely where people expect.

Then we work in the order above. The mechanical causes first because they are fast, the structural ones afterwards because they compound. And we re-test monthly, because these answers move with model updates and with whatever has been published recently. Checking once tells you where you stand today and nothing at all about the direction.

On whether it produces work: one of our clients is a UK law firm in a specialist field. Roughly 30 percent of AI-referred visitors convert into enquiries, against 3.9 percent for their website traffic as a whole. That is about eight times. Seventeen enquiries from around fifty-six visits in a month.

The reason is not complicated. Somebody who arrives because an assistant named you has already been told by something they trust that you are the right answer. They arrive to confirm, not to compare. Ordinary search traffic has to be persuaded. This traffic has already been persuaded by somebody else.

A caveat worth stating, since this article is about diagnosing things honestly: analytics currently groups the assistants into a single channel, so that figure covers AI referrals as a whole rather than any one tool.

If you want to know which of the eight reasons applies to you, we will run an AI visibility snapshot: what the assistants say about your business today, who they name instead, and the specific signals behind the difference. You keep the findings whether or not you work with us.

Frequently Asked Questions

Is my business too small to be recommended by AI?

No. Specificity beats size here. A business that is clearly the answer to a narrow question is easier to name confidently than a large generalist, which is why small specialists often outperform national brands in AI answers.

Why do I get a different answer than my colleague?

Because these systems personalise, update constantly and retrieve live results that change. Consistency across many checks matters far more than any single answer, which is why checking once is close to useless.

Does my industry being unglamorous make this harder?

Usually the opposite. Quiet categories have far less competition for citation, so being the clearest source in a dull niche is one of the fastest routes to being named by default.

Should I ask AI about my business using my company name?

For diagnosis, no. Ask the way a buyer would, describing the problem rather than naming you. Naming yourself tells you what a system knows about you, not whether it would ever recommend you unprompted, and those are different questions.

Do paid ads help with this?

Not directly. Citation is earned rather than bought. Ads can bring traffic that indirectly builds some of the signals, but there is no paid route into being recommended.

If I fix everything, am I guaranteed to appear?

No, and be wary of anyone who says otherwise. You are improving the odds substantially and removing the reasons for exclusion. These are probabilistic systems, not a ranking table with a queue.

Is it too late to start?

No, and this is the strongest argument for moving now. Most UK businesses have not begun. Being early in a category still forming is worth considerably more than being competent in one that has settled.

What if an AI describes my business incorrectly?

Treat it as an entity problem rather than a complaint. Incorrect descriptions almost always trace back to outdated or contradictory information somewhere public. Find it, correct it at source, and be patient, because these systems take time to catch up with changes.