Why Your Business Does Not Show Up in AI Search

That is the answer. The rest of this page helps you work out which of the five is yours, using checks you can run in an afternoon. No retainer, no phone call.

One honesty note about that table. The ordering is what we find repeatedly when auditing Ontario businesses that arrive with this complaint: a practitioner's frequency ranking, not a measured distribution from a controlled study. Nobody outside OpenAI, Google and Perplexity could produce one, because the retrieval layers are not published. Treat it as a search sequence, not a law.

The prerequisite Google states outright

Before any of the five causes matter, one condition has to hold for Google's surfaces. Google's AI optimization guide puts it in a sentence: to appear in generative AI features on Google Search, a page must be indexed and eligible to be shown with a snippet, meeting the standard technical requirements.

Read the second half again, because it catches people. Snippet eligibility is a separate switch from indexing. A page can be indexed perfectly and still be barred from AI Overviews because someone set a nosnippet directive, a max-snippet:0 robots meta tag, or a data-nosnippet attribute around the content you want quoted. Many were added during the 2024 and 2025 arguments over AI summarization and never removed.

Clear that gate first.

Cause 1: You have no entity footprint anywhere except your own website

This is the most common cause by a wide margin, and the one owners resist most, because the fix is not on their website. An assistant asked to recommend a business is running a verification problem, not a popularity contest. It needs corroboration: sources that are not you, saying you exist, do this work, and operate here. A site that is the sole publisher of every fact about itself gives it nothing to check.

Retrieval systems behave like a careful researcher who cannot confirm a claim: they reach for a competitor they can verify. We unpack how they assemble that evidence in where AI engines actually read from.

Fixing this is slow and unglamorous: trade associations, supplier and dealer listings, local chambers, industry directories that predate the AI era, local press, and review platforms where the review text describes what you do. None of it is a trick. It is the ordinary work of being a documented business, which most owners skipped because Google rankings never required it.

Cause 2: Your name, address and phone number disagree across the web

The second cause is a subset of the first and worse, because contradiction damages you more than absence. Three sources giving three different phone numbers is not three data points. It is a reason to lower confidence in all three and reach for a business whose record is clean.

There is a useful number here, with an equally useful caveat. SOCi's 2026 Local Visibility Index, published in January 2026 across roughly 350,000 locations and 2,751 multi-location brands, measured business profile accuracy at 68 percent on both ChatGPT and Perplexity, against 100 percent on Gemini, which draws on Google Maps directly. It found those brands in Google's local 3-pack 35.9 percent of the time, but recommended by ChatGPT only 1.2 percent.

Where this bites hardest is the detail nobody audits. A suite number on your website but missing from your Google Business Profile. A tracking number that appears nowhere else. A legal name that differs from your trading name everywhere else. Each is defensible alone. Together they make you look like several partially-described businesses instead of one well-described one. We covered the local case in AI recommendations versus Google Maps.

Cause 3: Your pages describe your business but never answer anything

Third cause, and the one most within your control this month. Most small business websites are brochures: who we are, what we value, why we care. Assistants are not answering "who are you". They are answering "who should I hire for X in Y", and they build that answer from passages they can lift. If no sentence on your site survives being pulled out of context, they have nothing to use.

This is where the academic work helps rather than decorates. Aggarwal and colleagues published the paper that named this discipline, GEO: Generative Engine Optimization, in November 2023. Testing against their GEO-bench benchmark, they reported visibility gains of up to 40 percent, strongest when content added quotations, statistics, and citations to authoritative references. Effect sizes varied by domain.

Worth stating what that does not mean. It is a 2023 paper measuring specific engines at a specific moment, and those engines have been rebuilt since. Nobody can promise you a 40 percent lift. What it supports is narrower and durable: verifiable specifics get selected more often than adjectives.

Note what Google says about writing this content, because it contradicts much of what is sold as AI optimization. Its guide states you do not need to write in a specific way just for generative AI search. There is no secret register, only whether your page holds checkable information. For the model behind the tactic, see what GEO actually is.

Cause 4: You are blocking the crawlers that matter, and probably not the ones you think

Fourth cause. Less common than the first three, but the only one that can take you from fully visible to fully invisible overnight, and the most misdiagnosed. Each AI company runs several crawlers with different jobs, and blocking the wrong one has consequences nothing like what people expect. Everything below comes from first-party documentation, not inference.

Three consequences fall straight out of that table, and each contradicts advice we see published weekly.

First, blocking GPTBot does not make you invisible in ChatGPT. It is a training opt-out. Pages telling you to unblock GPTBot so ChatGPT will recommend you are handing you the wrong lever. Blocking GPTBot while allowing OAI-SearchBot keeps you out of training and fully visible in search.

Second, blocking Google-Extended does not hurt your Google Search or AI Overview presence. Google says so in its own crawler documentation. Publishers who blocked it in 2024 fearing ranking damage were guarding against a risk the documentation says does not exist.

Third, robots.txt is not the whole story. Two of the seven agents above are user-triggered fetchers that may bypass it entirely. Your real blocking may sit in a layer you never think about: a Cloudflare bot rule, a WAF ruleset, or a security plugin quietly returning 403s to unfamiliar agents. Robots.txt can be immaculate while your firewall refuses the connection.

Of the five causes this is worth checking first despite ranking fourth, purely because it is cheap. Twenty minutes of reading configuration files either eliminates it or hands you an immediate fix.

Cause 5: The domain is genuinely too new to have a record

Fifth and least common, but real, and it deserves an honest hearing rather than a sales answer. If you registered your domain a few months ago, nobody has written about you, and no directory carries you, there is no diagnostic mystery. The systems are behaving correctly. There is nothing to retrieve.

Two things follow. First, no amount of on-page work compresses this to zero. Corroboration requires other people, who move on their own timeline. Any agency promising to make a two-month-old domain a cited authority by next quarter is selling something it cannot deliver.

Second, and more encouraging: a foundation repair company we worked with started from a fresh domain with no search history, no reviews and no lead flow, and reached 30 leads inside the first three months (internal lead tracking, 2026). That is lead volume, not traffic, and it came from doing the entity and content work in the right order. Lead volume varies by trade and season; foundation repair is high-ticket, so 30 enquiries is a material pipeline there in a way it would not be everywhere. Our Results Guarantee covers existing domains only, because a new domain's timeline is not something we can honestly commit to.

What is documented, what is inferred, and what nobody knows

The AI visibility field is full of confident claims with no basis behind them, so here is which category each thing on this page belongs to.

Documented, from first-party sources. Everything in the crawler table. Google's statements that snippet eligibility gates AI feature eligibility, that structured data is not required for generative AI search, and that it does not use llms.txt or any other machine-readable file you might be told to create.

Measured by third parties, with limits. The SOCi figures, valid for multi-location brands and not transferable to independents. The GEO paper findings, valid for the engines tested in 2023. Evidence, not laws of nature.

Genuinely unknown. The ranking function. How an engine weighs one corroborating source against another, its confidence thresholds, how often it refreshes its picture of you. None of that is public. We infer from observed behaviour and revise when the behaviour changes.

Recent work shows how unstable that layer is. Seo and colleagues published Verified Misguidance: Measuring Structural Citation Failures in Search-Augmented LLMs in May 2026, using 11,200 queries and 112,000 responses across ten models. They found 30.6 percent of citations distorted their sources. The finding most relevant to you: provider-level factors explained 88 to 96 percent of the variance in citation quality.

Running the whole diagnosis in one sitting

Sequence matters, and it is not the frequency ranking. Go cheapest first, so you do not spend three weeks on content while a firewall rule quietly refuses every crawler that matters.

Two failure modes to watch for. The first is fixing Cause 3 because writing is pleasant and the off-site work in Cause 1 is not. The second is deciding you are fine because one assistant answered once. Ask several engines, several times, phrased several ways.

Frequently Asked Questions

Written by Matt Griffin, founder of Formative Digital, Brantford, Ontario. Published 2026-07-20.