AI search news ·
Do AI engines cite homepages? In 1,147 cited URLs, 92.6% pointed at deep pages — and Claude cited zero homepages at all
92.6% of the 1,147 URLs that ChatGPT, Perplexity and Claude cited on our 12 small-business buyer questions (2026-08-03 run) pointed below the homepage. Claude cited zero homepages in 178 citations; Perplexity 17 of 717. ChatGPT is the one exception at 27% — and even there the homepages appear almost exclusively on tool-discovery questions, where the cited front page is a vendor's own. If your AEO plan is polishing the homepage, three of three URL-returning engines say the citable artifact lives deeper in the site.
Primary source: AskedAbout — URL-path classification of all full citation URLs in the four-engine buyer-question corpus, run of 2026-08-03
Method, stated before the numbers
Same corpus as the rest of the citation-economy series: 12 pinned buyer questions × 4 engines × 3 samples each = 144 answers in one run (2026-08-03), citations read from each engine's returned citation payload. The cut is mechanical: parse every cited URL and classify its path — homepage (empty path, e.g. `vendor.com/`) or deep page (anything below it). One exclusion, stated up front: Gemini returns bare domains in its citation payload — 440 entries with no path at all — so it cannot enter a path cut in either direction. That leaves the three URL-returning engines: ChatGPT (252 URLs), Perplexity (717) and Claude (178) — 1,147 full URLs across 348 distinct domains.
The split: three engines, one direction
| Engine | Cited URLs | Homepages | Deep pages | Homepage share |
|---|---|---|---|---|
| Claude | 178 | 0 | 178 | 0.0% |
| Perplexity | 717 | 17 | 700 | 2.4% |
| ChatGPT | 252 | 68 | 184 | 27.0% |
| All three | 1,147 | 85 | 1,062 | 7.4% |
| Gemini | 440 (domain-only) | — | — | excluded — payload carries no paths |
The deep pages are not scattered randomly through sites either: 52.7% of the 1,147 URLs sit in a recognizable content section (`/blog`, `/learn`, `/guide`, `/resources`, `/articles`, `/insights`, `/news`), and 34.0% have listicle-shaped slugs (containing “best”, “top-N”, “tools” or “-vs-”). And Perplexity's 17 homepage citations are not a diverse set: they cover seven domains, every one an AI-visibility tool vendor (aifindmybusiness.com, getvisibai.com, aivisitaudit.com, aiclicks.io, visibleiq.ai, trackmybusiness.ai, ailocaltool.com).
The ChatGPT exception is really a question-type effect
| Buyer question asked | ChatGPT homepage citations | Homepage share |
|---|---|---|
| “What tools track whether AI recommends my business?” | 20 of 27 | 74% |
| “What AI visibility tool works for a local business like a dentist or law firm?” | 20 of 26 | 77% |
| “Which tools show how ChatGPT, Perplexity and Gemini recommend businesses?” | 9 of 22 | 41% |
| “How can I check what ChatGPT says about my business?” | 0 of 16 | 0% |
| “How do I improve my business's visibility in ChatGPT and AI search?” | 0 of 24 | 0% |
| “What is a good AthenaHQ alternative for a small business?” | 0 of 19 | 0% |
ChatGPT's 27% average hides a clean split: when the question asks which tool exists, it cites vendor front doors — 74–77% homepages on the two pure tool-discovery questions. When the question asks how to do something, its homepage rate collapses to zero and it cites articles like the other engines. The homepage citations are not “homepages beat content” — they are ChatGPT naming products, where the product's address happens to be its homepage.
What this means if you care about AI visibility
- Your homepage is almost never the citable artifact. On advice-shaped and how-to buyer questions — most of what a small business's customers actually ask — all three URL-returning engines cited deep pages essentially 100% of the time. The unit of AI visibility is a specific page that answers a specific question, and it usually lives in a content section.
- The homepage path only pays if you ARE the answer. The homepages that earned citations were tool vendors being named on tool-lookup questions. That is product-in-category visibility, not content strategy — you get it by being a nameable vendor in the category, not by rewriting the front page.
- This is the same shape a 25,000-citation agency study found from the other side. DeltaV's 90-day study measured citation “fingerprints” by industry — listicles 61% in B2B tech, homepages 55% in local services. Our panel is a B2B-ish tool category, and our first-party numbers land where their fingerprint predicts: deep listicle-and-article pages dominate, homepages only surface where the vendor itself is the answer.
Limits
- One run, one day, one panel. 12 buyer questions about AI visibility for small businesses, 2026-08-03, API citation payloads — consumer UIs can differ. The full URL list behind every number is dated and shared with anyone who asks.
- “Homepage” means an empty URL path. A root URL with a query string counts as a homepage; `vendor.com/index.html` would count as deep. With 85 homepage rows in 1,147, every classification is checkable in seconds.
- Gemini is excluded by its payload format, not by choice. 440 domain-only entries carry no path information — a limitation we've also written about from the tagging side.
- We have a commercial interest. We sell AI-visibility audits; discount our conclusions accordingly and read the counted rows.
The practical reading: if AI engines are citing anyone in your category, they are citing a deep page — a comparison, a how-to, a listicle — not a front door. Whether any of those cited pages are yours is checkable directly: run the four-engine, three-sample measurement on your business in 60 seconds.
See your number
See which businesses AI names when your client's buyers ask.
Running this for clients? The $249 agency 5-pack audits five businesses, white-labeled.
Who runs this
- Built and operated by Sensara LLC, Atlanta, Georgia — about us and how the audit works.
- See what the report looks like before you run anything — score per engine, the competitors AI names instead of you, and a fix plan.
- We run the same audit on ourselves every week and publish the result: in the latest run AI named AskedAbout in 1 of 144 answers. We report our own numbers the way we report yours.