Check this page with an assistantOpens a chat asking it to summarise this article and name the evidence behind each claim.
Claude opens with the prompt on your clipboard: Anthropic does not support prefilled prompts on the web, and we would rather copy it than ship a button that drops it.
How Do ChatGPT, Perplexity, and Gemini Pick Which Businesses to Name?
Every major assistant answers commercial questions the same way: it runs a live search, pulls a pool of candidate pages, and composes an answer from the passages it trusts most. ChatGPT leans on Bing's index for browsing, Perplexity runs its own crawler, and Gemini and AI Overviews draw from Google's index. The business that gets named is the one that appears, consistently described, in the sources each engine retrieves.
This matters because it kills the most common misconception: that there is some registry to get into. There is not. When someone asks "what's the best bookkeeping service for restaurants," the assistant fans that question out into sub-queries, retrieves pages for each, and cross-references which names keep showing up with the same description. A business mentioned once, on its own site only, reads as unverified. A business described the same way on its site, its Google Business Profile, two directories, a comparison page, and a Reddit thread reads as consensus.
The practical unit of competition is not your homepage. It is the individual passage: a paragraph, table row, or heading-plus-answer block that can be lifted verbatim into a generated response. Sites win AI mentions passage by passage.
Gate 1 and Gate 2: First You Get Retrieved, Then You Get Cited
Think of AI visibility as two gates. Gate 1 is entering the retrieval pool: the engine has to pull your page as a candidate at all, which is won with topical relevance, entity coverage, and crawlable structure. Gate 2 is citation selection: among retrieved pages, the engine quotes the ones with clean, self-contained answer blocks. Failing either gate produces the same result, which is silence.
Most businesses fail Gate 1 for a mundane reason: the pages that would answer buyer questions simply do not exist. If nobody has written your pricing explanation, your comparison against the obvious alternative, or your "is this right for X" page, there is nothing to retrieve. This is why the fastest gains usually come from mapping which buyer questions have no page yet rather than from polishing the homepage.
Gate 2 failures look different. The page ranks, the information is in there somewhere, but the answer is smeared across four paragraphs of narrative. Extraction engines skip prose they cannot lift cleanly. The fix is structural: put a direct two-sentence answer immediately under each question-shaped heading, keep facts in tables, and make each section self-contained enough to stand alone out of context.
What Google Actually Says About AI Search (and the llms.txt Myth)
Google's published guidance for its AI features is unusually plain: foundational SEO still decides visibility, no special AI markup or llms.txt file is required, and scaled content produced primarily to game rankings violates its spam policies. Everything sold as a secret "AEO hack" contradicting those three statements is contradicting the vendor of the search engine itself.
While building SearchHandled's verification gate we went through the primary documents rather than the commentary around them, and three points from Google Search Central are worth quoting in substance. First, its guidance on appearing in AI features says the same helpful-content fundamentals apply, with no additional technical requirement. Second, its March 2024 spam policy update defines scaled content abuse as producing many pages primarily for rankings rather than readers, regardless of whether a human or an AI wrote them. Third, its starter guide de-emphasizes tricks like keyword-loaded domains in favor of clear structure and consistent naming.
The strategic reading: AI search did not replace the rules, it raised the price of ignoring them. Thin pages that once quietly ranked now also fail retrieval, while specific, well-structured pages get quoted in two channels at once.
The Six-Step Plan for Earning AI Assistant Mentions
The plan in one line: make your entity consistent everywhere, then publish extractable answers to the questions your buyers actually ask. Six steps, in priority order, each one checkable.
- Lock the entity. Use one exact business name, one canonical one-sentence description, and one URL across your site, Google Business Profile, LinkedIn, and every directory that lists you. Assistants cross-check descriptions; conflicting ones dilute confidence.
- Build the comparison pages.Name your real alternatives and compare honestly, tradeoffs included. Assistants love comparison content because it maps directly onto "which should I choose" questions. Our own compare pages exist for exactly this reason.
- Answer zero-volume questions. The specific questions from your sales calls and support inbox usually show zero volume in keyword tools, yet they are precisely what people type into chat interfaces. One focused section per question, direct answer first.
- Structure for extraction. Question-shaped headings, a two-to-three sentence answer directly beneath each, real HTML tables for anything numeric. If a passage cannot be quoted alone, it will not be.
- Earn third-party corroboration. Reviews on platforms assistants retrieve (Google, Trustpilot, G2 for software), a presence in relevant Reddit and community threads, and consistent mentions elsewhere. One self-hosted claim is an assertion; the same claim in three places is an answer.
- Keep pages visibly current. Dated updates, current-year pricing, and refreshed statistics. Retrieval favors fresh sources for commercial queries, which is the same reason refreshing old posts outperforms only writing new ones.
Where Each AI Surface Pulls From, and What That Means for You
| AI surface | Primary retrieval source | Highest-leverage move |
|---|---|---|
| ChatGPT (browsing) | Bing index plus cited web pages | Verify your site in Bing Webmaster Tools and fix indexing there, not just in Google |
| Perplexity | Own crawler, weights structured pages and communities | Publish comparison tables and maintain honest community presence where your niche discusses vendors |
| Google AI Overviews | Google index, extracts passage-level blocks | Direct answers under question headings; facts in visible HTML tables, not only in JSON-LD |
| Gemini | Google index and Knowledge Graph entities | Consistent entity naming everywhere so the Knowledge Graph can resolve your business confidently |
Retrieval behavior as documented and observed mid-2026; engines change weighting continuously, sources rarely.
Note what is absent from that table: any lever you can buy. Every cell is content or infrastructure work on surfaces you already control or can participate in honestly.
The Cost Math: AI Visibility Against Paid Clicks
A services business paying a typical $4 cost per click on search ads buys 250 visits for $1,000, and the meter resets to zero every month. The same $1,000 spent producing four solid answer pages keeps serving visits, and AI citations, for years at zero marginal cost. The break-even usually lands within the first few months, and everything after is margin.
The comparison is not entirely fair in either direction. Ads deliver tomorrow; content compounds slowly and unevenly. But the asymmetry that matters for AI search is this: assistants do not read your ads. A business with zero organic footprint is invisible to the fastest-growing referral channel regardless of ad budget, which is why even ad-heavy businesses now need an answer-shaped content base. For a small team, the workable split is ads for this quarter's pipeline and extractable content for every quarter after, with free tooling or an automated pipeline keeping the content side from consuming your week.
Questions People Ask About AI Recommendations
- Why doesn't ChatGPT recommend my business when people ask for companies like mine?
Almost always because your business never enters the retrieval pool: the pages that would answer the question do not exist, are not indexed in Bing or Google, or describe your business inconsistently across the web. Assistants cannot cite what they cannot retrieve and verify. Fix existence and consistency first, then extraction structure; optimization tricks cannot compensate for either.
- Can you pay ChatGPT or Perplexity to recommend your business?
No. There is no submission form, listing fee, or ad product that inserts a business into organic AI answers. Assistants cite whatever their retrieval layer finds and trusts, so the only durable lever is being present, consistent, and extractable on the open web. Sponsored placements in some AI products are labeled ads and are a separate channel.
- Do you need llms.txt or special AI markup to appear in AI answers?
No. Google states directly that no special file or markup is required for its AI features and that foundational SEO practices are what count. An llms.txt file does not hurt, but treating it as a strategy is a distraction from entity consistency and extractable content, which are the signals that actually move citations.
- How long does it take to start appearing in AI assistant answers?
Pages can enter retrieval pools within days of being indexed, but brand-level consensus builds over weeks to months because assistants cross-check multiple sources before naming a business confidently. Comparison pages and specific long-tail answers typically earn citations fastest because they face the least competition.
- Does schema markup still matter for AI search visibility?
It helps machines confirm what a page says, but it cannot rescue content that is not extractable in the visible HTML. Put facts in headings, tables, and short answer blocks first, then mirror them in structured data. Google notes structured data is useful but is not a requirement for its AI features.
