If you're invisible in AI search, the cause is almost always one of five layers — and they stack in order. An engine has to reach your page, recognise you as an entity, extract a clean answer, find your claim corroborated, and see it's fresh enough to retrieve. Diagnose top-down: a brilliant page no crawler can fetch is invisible for a reason that has nothing to do with the writing.
Most "why am I not showing up?" panic comes from fixing the wrong layer — rewriting copy when the real problem is a blocked crawler. Here's the ladder.
Why am I invisible in AI search?
Because you're losing at one of these five layers, checked in this order:
| # | Layer | The question it answers | If you fail here… |
|---|---|---|---|
| 1 | Access | Can the engine fetch your page at all? | You're not in the index. Nothing else matters. |
| 2 | Entity | Does it recognise you as a name for this topic? | Your page is read but not trusted enough to cite. |
| 3 | Extractability | Can it lift a clean answer from the page? | Your facts are there but unquotable. |
| 4 | Corroboration | Does anything else confirm your claim? | You read as a lone, self-interested assertion. |
| 5 | Freshness | Is the page recent enough to retrieve? | You've decayed off the citation cliff. |
Work it from the top. Each layer assumes the ones above it are clear.
Layer 1: Can AI engines even reach your pages?
The most common and most fixable cause. Content an engine can't fetch can't be cited — full stop. Three things block access:
robots.txtrules that disallow AI user-agents likeGPTBot,ClaudeBot, orPerplexityBot.- CDN-level blocks — many CDNs now challenge or deny AI bots by default. This is the silent one; see is your CDN blocking AI crawlers.
- JavaScript-only rendering — if your facts only appear after scripts run, many crawlers never see them.
Confirm what's actually reaching you in how to see which AI bots crawl your site, and understand the pipeline in the AI crawler lifecycle. Until access is clean, skip the other four layers.
Layer 2: Do engines know you as an entity?
Engines lean heavily on how established you are as an entity — your brand's search volume, knowledge graph presence, and how often you co-occur with the topic across the web — often more than on backlinks. A page from a recognised name inherits authority; an identical page from an unknown one doesn't.
This is the slow, durable lever. Be consistent and corroborated across the web so the model associates your name with the topic. The mechanics of why this decides citations are in why AI cites one brand and ignores a near-identical competitor and how AI engines choose which brands to recommend.
Layer 3: Can a model extract a clean answer from your page?
Retrieval happens at the passage level, not the page level — a model lifts one self-contained chunk. If your answer is buried mid-paragraph or diffused across the page, it's effectively unquotable even when it's correct.
The fixes are structural: open each section with the answer in ~40–60 words, use question-style headings, put facts in lists and tables, and keep chunks self-contained. This is the same discipline as accessibility — accessibility and AI-parseability — and the craft is laid out in how to get cited by AI.
Layer 4: Is your claim corroborated anywhere else?
Engines favour claims echoed by multiple independent, authoritative sources. A figure repeated on Reddit, review sites, and the press reads as corroborated and safe to cite; the same figure sitting only on your own site reads as a lone, self-interested assertion.
So earned presence does work a brand page can't: third-party best-of lists get cited far more than self-listing pages for commercial queries — pursue placement in them rather than writing your own (see how to get into the best-of lists AI engines cite). And keep it honest: trying to fake corroboration backfires, as we cover in can AI search be manipulated.
Layer 5: Is your page fresh enough to be retrieved?
Live retrieval favours recently-updated pages, and citations decay after roughly a quarter — the three-month citation cliff. A "publish and forget" page slides out of answers even if it once appeared. Put competitive pages on a refresh cadence with substantive updates (new data, corrected facts), not date-bumps. The evidence and cadence are in the content freshness citation cliff.
One caveat that prevents a false alarm: there's a documented crawl-to-cite lag of weeks. Training and search bots often crawl a new page heavily before the index starts citing it, so heavy crawling with zero citations is a leading indicator, not a failure — don't judge a page dead in week two.
How do you tell which layer is the problem?
Don't guess — test across engines and read the pattern:
- Check access for the URL:
robots.txt, CDN response to AI user-agents, and whether facts are in the raw HTML. - Run the same prompt across ChatGPT, Claude, Perplexity, and Google's AI surfaces; log who gets cited.
- Read the pattern: absent everywhere points to entity, extractability, or corroboration; absent on some engines only almost always means access or index coverage on those engines.
- Check freshness and off-site mentions for the topic.
If the gap is consistent across every engine, it's content, entity, or corroboration. If it's specific to one or two engines, it's almost always crawler access or index coverage — the easier kind to fix.
That cross-engine comparison — the same questions, tracked over time, logging exactly who gets cited and where — is precisely what Buffy Intel does, so you can act on a diagnosis instead of a hunch. Start at the top of the ladder; you'll usually find the blocker before you reach the bottom.
