Absent From the Answer: What 33 AI-Search Audits Reveal About Brand Visibility

When buyers ask AI which business to choose, most brands are not in the room.

Over the second half of 2026 we ran the same experiment on 33 businesses. We stopped guessing how AI search treats them and simply asked — putting the real questions a buyer types into ChatGPT, Google’s AI Overviews, Perplexity, and Gemini, and recording, for each answer, whether the engine named the brand, mentioned it in passing, or left it out entirely.

We logged 177 buyer-intent prompts across those 33 brands. The pattern was consistent enough to call a rule:

82.5% of the time a buyer’s question was asked, the brand that answers it best was absent from the answer — or buried inside someone else’s.

Only 17.5% of answers named the audited brand at all. Just 5.1% put it first. These were not obscure businesses. Every one ranks in classic Google search, holds real credentials, and in several cases owns the category it invented. The engines simply could not see them where it counted.

This is not a story about weak brands. It is a story about a new layer of search that most brands have not yet been built for — and the good news buried inside the data is that the fix is unusually tractable.

Finding 1 — Being named is the exception, not the rule

We graded every one of the 177 answers on a four-point scale for how the engine treated the brand: Leads (first-named or recommended), Named (present and correct), Partial (cited, but not the one chosen), or Absent (not in the answer at all).

The single most common outcome, by a wide margin, was absence: the engine composed a confident, sourced answer to a question the brand exists to answer, and never mentioned it. The second most common was partial — the brand appears, but as a footnote inside a narrative written by review sites, directories, and competitors.

Together, those two states — invisible or subordinate — account for 82.5% of every answer we logged.

Finding 2 — The gap tracks the buying journey

Break the 177 prompts down by the kind of question, and the story sharpens. Brands are named where the question is about what they are — and vanish almost exactly where the question becomes which one to buy.

Now bundle the four question types that sit closest to a purchase — comparison, alternatives, product-fit, and price. That is 34 prompts where the buyer has stopped exploring and started choosing.

The audited brands were named in exactly zero of them. 0 of 34.

Every single time, the engine handed the “which one should I pick” and “what does it cost” answers to review sites, aggregators, and competitors.

This is the study’s spine. A brand’s own website is strongest at describing what it is, and the engines reward that — reputation and definitional questions are where brands score best. But the moments that actually move a sale (“X vs Y,” “best X for Y,” “X alternatives,” “how much does X cost”) are answered from comparative sources most brands never authored. No one wrote the comparison, so the engine used someone else’s.

Finding 3 — A third of brands are effectively invisible

The prompt-level pattern rolls up into a stark brand-level one.

• 12 of 33 brands (36%) were never once led or clearly named — across their entire prompt set, not a single clean citation.
• 5 more brands earned a citation only when the buyer typed their exact name, which is useless for discovery.
• That means 17 of 33 brands (52%) are either invisible in AI answers, or visible only to someone who already knows to look for them.

The brands that did break through share a tell: they win a narrow, defensible slice and lose everything adjacent. A life-science instrument maker leads “cell washing without a centrifuge” — a term it coined — yet is absent from its own alternatives query. A credit union is the first-named answer for “best credit union in Alabama” but missing from every rate-comparison query. Owning the vocabulary buys you the definitional question. It does not buy you the sale.

Finding 4 — It’s a plumbing problem, not a reputation problem

Here is the encouraging part. The same handful of missing signals recur across almost every brand — and they are all fixable with markup and content, not years of brand-building.

An answer engine builds a response from text it can lift, facts it can trust, and sources it can attribute. A claim trapped in a hero image is invisible to it. A testimonial with no review schema is unquotable. A price that lives only in a sales call cannot enter a “how much” answer. And with no brand-authored comparison, the “which is better” verdict defaults to whoever did publish one.

Two things we expected to matter — and didn’t

A study is only as honest as the results it is willing to report against itself. Two of ours came back null:

• llms.txt is not a magic bullet. Brands that had published one were cited at the same rate as brands that hadn’t — 18% vs 18%. It is table-stakes hygiene, not a lever. What moves citations is extractable proof, schema, and off-domain presence.
• The gap is not a B2B or B2C phenomenon. Win rates were identical across both — 18% vs 18%. This is not about audience. It is about publishing structure.

Closing the answer gap

Every recommendation below is drawn directly from what separated the brands that got cited from the ones that didn’t.

1. Author the comparison you keep losing. Write the “vs,” “alternatives,” and “best X for Y” pages for your own category — honest, specific, table-driven. This is the 0-of-34 zone: unowned, and it decides sales.
2. Make every proof point extractable. Pull your numbers, credentials, and differentiators out of images, PDFs, and JavaScript and into plain, quotable on-page text — ideally a 40–60 word lead answer under each buyer question.
3. Mark up what you already have. FAQPage, Review/AggregateRating, Product, and a complete Organization schema — the assets 79–88% of brands were missing. The content usually exists; only the machine-readable layer is absent.
4. Put the price on the page. A range, a starting point, a “from” figure — anything machine-readable. Price questions were lost 100% of the time, almost always to third parties guessing on the brand’s behalf.
5. Fix the entity off your own domain. Reconcile the review sites, directories, and databases the engines quote. Where an aggregator lists the wrong competitors or a split identity, that off-domain record — not your homepage — is writing your answer.

The takeaway

Ranking got you found. It no longer gets you chosen.

The businesses in this study were not beaten on merit. They were beaten on structure — absent from the exact answers where buyers decide. The signals that fix it are near-universally missing, low-cost, and fully within a brand’s control. The window is now, while the answers are still being written and the incumbents are still absent too.

Methodology

This study aggregates 33 first-party AI-search-visibility audits conducted by Digital Elevator during the second half of 2026.

The sample. We chose brands for spread, not convenience: manufacturers and med-tech, SaaS and professional services, e-commerce, hospitality, local clinics, and a credit union; a family cabin rental and a global central lab. Of the 33 brands, 17 are B2B, 15 are B2C, and 1 is a hybrid; 24 are U.S. businesses and 9 are international or global. By business model the set breaks down as professional services (11), then manufacturers, SaaS, local services, and hospitality (4 each), e-commerce (3), and others. If a gap shows up across all of that, it is structural — not a quirk of one industry.

The questions. For each brand we wrote a basket of prompts spanning the buying journey — the definitional “what is” question, the category question, the head-to-head comparison, the “alternatives” question, the “best X for Y” shortlist, the price question, the local “near me” question, and the brand-name reputation check. A median of five to six prompts per brand (177 in total), phrased as a real buyer would type them — in the brand’s own language, including Portuguese, German, and Chinese where the market demanded it.

The engines. Prompts were run against the AI answer surfaces buyers actually use: ChatGPT, Google’s AI Overviews, Perplexity, and Gemini, plus the third-party sources those engines draw on.

The verdict. Each answer was analyst-graded on a fixed four-point rubric — Leads, Named, Partial, or Absent — for how the engine treated the brand. We also recorded which sources and competitors did win each answer, and the on-page and off-page signals behind the result.

Limitations. AI answers vary by phrasing, personalization, and time, so these figures describe this corpus of 33 audits, not a probability sample of all businesses. Prompt baskets were tailored per brand rather than held identical, a deliberate choice that favors representativeness of each brand’s real buying journey over uniform wording. Grading involves analyst judgment against a fixed rubric. Where results ran against our own priors — the llms.txt and B2B/B2C nulls — we report them alongside the positive findings.

Full per-brand records are available on request.

---

Digital Elevator helps brands get named in the AI answers their buyers trust. If you’d like to see where your own brand stands, request an AI visibility audit.