A founder spent months building traditional SEO for a fintech brand — ranking on Google, generating organic traffic — then noticed something troubling: their content was invisible to Perplexity, ChatGPT, and Google's AI Overview. The Google Search Console numbers looked fine. The AI tools cited competitors instead.
That gap is showing up across verticals, and most agency clients have no idea it exists.
How AI Search Actually Works
AI search tools don't operate on Google's crawl-and-rank model. Perplexity, ChatGPT Search, and Google's AI Overviews pull from a mix of real-time web access, pre-training data, and curated high-trust sources. These models aren't ranking ten blue links — they're synthesizing an answer and citing the sources they trusted enough to quote.
Traditional SEO optimizes for crawl coverage, keyword density, and backlink authority. AI search optimizes for something different: can the model confidently attribute a specific fact to your page without hallucinating?
A page ranked #3 for "best fintech savings account rates" may be optimized for keyword matches and click-through rate. That same page might never appear in a ChatGPT response for "which savings account works best for a gig worker?" — because the answer isn't clearly extractable in two sentences, and the model can't cite it cleanly.
Why Your Google Rankings Don't Transfer
The signals that move the needle in AI search are different from traditional SEO signals, and several of them are invisible to standard audits:
- Structured data (JSON-LD schema): AI models read schema markup to understand entities — your business type, products, FAQ answers, review scores. A page without schema forces the model to infer context from unstructured prose. It often gets it wrong, cites something else, or skips the page entirely.
- Answer placement: AI models extract the first clear declarative statement on a page. Burying the answer in paragraph four is fine for SEO copywriting. For LLM citation, you need the direct answer at the top, before the preamble.
- E-E-A-T signals: Google's Experience, Expertise, Authoritativeness, Trustworthiness guidelines were always important — they've become more important as AI Overviews pull from verified, high-trust sources. Author bios, publication dates, and clear entity pages (About pages with structured credentials) all factor into whether a model treats your content as citable.
- Thin and near-duplicate pages: Site architectures that generate hundreds of similar pages for product variations, ZIP codes, or filtered categories worked for long-tail SEO. AI search collapses those into a single representative source — or ignores them. The same pattern that caused WordPress to create scale problems at the content layer is now creating AI-visibility problems at the citation layer.
AI Crawlers Are Already on Your Clients' Sites — Are They Blocked?
Most site owners never updated their robots.txt after the AI crawler wave started. GPTBot (OpenAI), PerplexityBot, ClaudeBot (Anthropic), and Google's AI crawlers are all requesting access to pages daily. A blanket Disallow: / rule added during the panic period of early LLM scraping debates may have blocked all of them.
Pull the robots.txt for any client where AI search visibility matters. Check whether those crawlers are explicitly disallowed. If they are, decide deliberately whether to allow them — not based on the default setting from three years ago.
On the monitoring side, Bing's AI Performance Dashboard is currently the most accessible benchmark for how often your content appears in AI-generated responses. Google's equivalent data is less transparent — you can infer AI Overview appearances from a drop in position-zero clicks in Search Console, but there's no direct reporting yet. Start with Bing to establish a baseline.
The llms.txt Standard: Worth Knowing, Not a Fix
An emerging convention called llms.txt (analogous to robots.txt) lets site owners signal which content they want AI models to index and cite. The format provides plain-text summaries of key pages, making content easier for models to parse without running full HTML processing.
It's worth implementing — but it doesn't fix a site that lacks structured data, clear entity definitions, or direct-answer content. Think of it as a layer on top of a solid technical foundation. If the foundation is thin pages and unstructured prose, an llms.txt file just makes the thin content easier to ignore faster.
Adoption is also early. Not all models honor the spec, and there's no enforcement mechanism. Implement it because it's low-effort and directionally correct, not because it solves the problem.
The Audit Your Clients Actually Need
A traditional SEO audit and an AI-search audit ask different questions. Before billing for SEO coverage, agencies need to run both. The AI-search version:
- Can a model identify what this business is and does in one sentence? Check the homepage, About page, and key service pages for clear, machine-readable entity definitions. If a model reads the homepage and can't extract a clean business description, neither can AI search.
- Does the content answer the five questions customers actually ask? Take the top five queries driving traffic. Find the answer on the site. If it takes more than two sentences to extract it, restructure. Direct answers near the top of the page, before supporting context.
- Are AI crawlers blocked? Check
robots.txtfor GPTBot, PerplexityBot, ClaudeBot, and Google-Extended. Confirm the allow/disallow state is intentional. - Is schema markup implemented on the key pages? Organization schema on the About page. Product or Service schema on service pages. FAQ schema on any page that answers questions. These are table stakes now, not enhancements.
- Are there thin or near-duplicate pages that need consolidation? Canonicalize or restructure before AI crawlers build a low-confidence picture of what the site is actually about.
The Timeline Problem
This isn't a crisis arriving someday. Fintech, legal, healthcare, and local services are already seeing the shift — users research with Perplexity and ChatGPT first, then click through. A client's Google rankings can look stable while their AI search presence is zero, and they won't see it in standard monthly reporting because no one is measuring it yet.
The agencies adding AI-search visibility to their baseline audit now are building that capability before clients start asking. The agencies that wait until a client notices the traffic pattern will be retrofitting this across an entire portfolio at once.
Start with the robots.txt check and the Bing AI Performance Dashboard. Both are free, both take under an hour per client, and both will surface whether there's a problem worth fixing.
