Why isn't AI recommending my business?

You typed it in yourself. Something like best accounting firm for construction companies in Denver, or whatever the exact phrasing is in your business. ChatGPT named three companies. None of them were you, and one of them you have out-serviced and undercut for six years.

Worth taking seriously, and worth not panicking about — one answer to one question is very weak evidence, for reasons the measurement section covers. But if the pattern holds across several real questions, the cause is almost always one of six things, cheapest first.

Six causes, in the order worth checking them

The two mechanical causes: blocked, or absent from the index

Rule these out first. They are binary — either the engine can reach you or it cannot, and no writing fixes a Disallow line.

One: you blocked the crawler that decides citations

OpenAI runs several crawlers doing different jobs. GPTBot collects content for model training. OAI-SearchBot crawls so pages can be cited in ChatGPT search, and OpenAI's own documentation states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. Two robots.txt tokens, two entirely different outcomes, and the names do not make that obvious.

At a lot of companies, someone decided in 2023 or 2024 that they did not want AI training on their content, and a developer implemented it as a blanket block on anything with OpenAI's name attached. The decision was about training. The cost landed on search, and nobody noticed.

# BAD — removes you from ChatGPT search answers
User-agent: OAI-SearchBot
Disallow: /

# ALSO BAD — a wildcard block catches OAI-SearchBot too
User-agent: *
Disallow: /

# WHAT MOST BUSINESSES ACTUALLY WANT:
# citable in answers, excluded from training
User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: bingbot
Allow: /

User-agent: GPTBot
Disallow: /

User-agent: Google-Extended
Disallow: /

Sitemap: https://example.com/sitemap.xml

Two: you are not in the index that engine reaches for

There is no single AI index. Each assistant reaches for a different retrieval layer, and being strong in Google tells you little about the others.

Necessary, but not sufficient. One vendor analysis found Claude's cited results matched Brave's top organic results about 86.7% of the time, against roughly 26.7% alignment between ChatGPT and Bing — ChatGPT re-ranks far more aggressively before deciding what to quote. Single-vendor figures, so treat them as directional.

Cause three: you have no page for the question

The most common content failure, and always the same shape. The customer asked for the best bookkeeping software for a two-person law firm. Your site has a page called Solutions. A page cannot be cited as the answer to a question it never asks.

Ahrefs analyzed 1.4 million prompts against ChatGPT desktop data in February 2025, comparing pages that were cited against pages that were not. Date that study when you use it — it predates a year of ranking changes — but the core finding has held up.

Strip out the decimals and it is almost embarrassingly plain: the page that gets cited is the one whose title is closest in meaning to the question asked. Not the page with the most backlinks. Not the longest page.

• One question per page, stated in the title. If a reader would not type your H1 into a chat box, it is not an answer page.
• The answer in the first sixty words. Assume an engine lifts that paragraph and nothing else.
• Readable slugs. /best-bookkeeping-software-for-small-law-firms carries meaning that /services/item-4471 does not.
• Specifics worth quoting. Named tools, real numbers, dated studies, actual prices. Generic advice is unquotable.

The work is not writing more. It is writing the ten pages that match the ten questions a customer asks in the week before they buy, and naming each one after its question.

Cause four: nobody else talks about you

This is the hard one, and it is why a technically flawless site can still lose. When someone asks for the best option in a category, the pages that serve that question well are comparative — roundups, directories, review sites, forum threads, trade press. Your own homepage is a weak source for a comparative question, because it has an obvious interest in the outcome.

One aggregator analysis put roughly 84% of AI citations on earned media rather than brand-owned pages, with only 5% to 10% from the brand's own website. That comes from a marketing-industry aggregator, not a platform or a peer-reviewed study, so treat the exact number with suspicion. The direction matches better-sourced work: Ahrefs found the overwhelming majority of ChatGPT citations were ordinary search results, dominated by third-party pages.

Fixing it is slow. It means being included in the roundups and directories that already rank for your category, getting quoted in trade publications, and being present where your category is discussed. There is no shortcut, which is why it is durable — a competitor cannot undo six years of other people mentioning you.

Cause five: the engine is not certain who you are

An answer engine has to decide that the Acme Plumbing in a Denver directory, the Acme Plumbing & Heating LLC on file with the state, and the Acme Plumbing Denver on a reviews site are one company. When your name, address, and description vary across the web, that link weakens — and a system unsure who you are will not confidently recommend you.

The Wikipedia evidence here is real but frequently oversold. A 2025 cross-LLM study covering four major models and 58 questions found 50% of the marketing agencies most frequently cited in AI answers had a Wikipedia page. That is a correlation with an easy non-causal explanation: an agency notable enough to sustain an article is also large enough to be written about everywhere else.

The caution applies harder to Wikidata. GEO vendors assert a Wikidata entry acts as a truth source AI systems consult. No AI operator has confirmed that mechanism publicly and no peer-reviewed work supports it. Keeping an existing entry accurate is cheap. Building a strategy on it is not supported by anything currently public.

What is worth doing costs an afternoon: one exact company name used identically everywhere, one address and phone format across every listing you control, and plain text on your site stating what you do, who you serve, and where. Ambiguity is the problem. Consistency is the fix.

Cause six: nothing is wrong and the category is unsettled

Before spending a quarter on this, check whether there is anything to win. Semrush and Kevin Indig studied 1,094 categories, more than 50,000 brands, and over 600,000 citations from ChatGPT data between January and June 2026.

The same study defines what owning a topic means: appearing in at least four of five related prompts, with a five-point lead over the next brand. Narrower wins reversed frequently. The competitor who showed up in the one query you ran this morning may not be there next week.

How to measure this without fooling yourself

The instinct that brought you here — asking ChatGPT about yourself once — is a genuinely bad test, for three reasons that compound.

1. A sample size of one against a volatile system. Semrush tracked over 100 million citations across 230,000 prompts for 13 weeks in 2025 and watched ChatGPT's Reddit citation share fall from roughly 60% to about 10%, and Wikipedia from around 55% to under 20%. Perplexity and Google's AI Mode stayed comparatively stable. ChatGPT did not.
2. Your own session is contaminated. Personalization, memory, and chat history all shape what comes back. The answer you saw is not the answer a stranger sees.
3. You are scoring the wrong thing. Rank is a search-era metric. In an answer you are either named or you are not.

A routine that tells you something looks like this. Write down 10 to 20 real buying-intent questions — the ones customers ask in the week before they choose, not the ones about your brand name. Run each across several engines: ChatGPT, Google AI Mode, Perplexity, Claude, Copilot. Use a fresh or logged-out session. Log results the same day each week or month. Score presence, not position: were you named at all, in how many questions, on how many engines. Then read the trend, not any single cell.

For first-party evidence, ChatGPT appends utm_source=chatgpt.com to its outbound links, so cited pages that get clicked show up in analytics. Understand what that undercounts. Most AI answers end without a click — the SparkToro and Similarweb clickstream study of US search behavior from January to April 2026 measured 68.01% of Google searches ending with no click at all, and an answer engine ends the session more decisively still. Referral counts are a floor on how often you are cited, not a measure of it. That gap is why GetFound3 scores presence across engines rather than referral traffic.

What success actually looks like

One reality check, because the expectation matters more than the tactics. Fixing all six of these will not produce a traffic flood, and any post promising one is selling something.

Chartbeat data published via Axios in March 2026 found chatbot referrals still accounted for less than 1% of total publisher pageview referrals — despite ChatGPT referral traffic growing more than 200% during 2025. Over the same period Google Search referrals to publishers fell 34% year over year, with small publishers down 60%.

AI is eating Google's clicks faster than it is replacing them.

That changes what you are optimizing for. The value of being named is not the click. It is that a buyer who hears your name will search it directly, ask a colleague, or type it into their browser — none of which lands in analytics as an AI referral. The goal is being the recommendation. Harvesting the click is a secondary and shrinking prize.

The checklist

1. Read your robots.txt. Confirm OAI-SearchBot, PerplexityBot, Claude-SearchBot, and bingbot are allowed. Decide GPTBot and Google-Extended deliberately.
2. Verify the site in Bing Webmaster Tools and submit your sitemap. That covers ChatGPT and Copilot at once.
3. Check three key pages on Brave and in Google Search Console. Confirm they are indexed, not just submitted.
4. Write down the ten questions customers ask before buying and mark which have a page. The gaps are your content plan.
5. On existing pages, move the answer into the first sixty words and give the URL a natural-language slug.
6. List every roundup, directory, and thread ranking for your category, and note which include you. That is your earned-media plan.
7. Standardize your company name, address, and phone across every listing you control. One format, everywhere.
8. Run those questions across three engines logged out, score presence, repeat monthly, and judge the trend.