AI search news ·
Does AI cite one page or many from the same site? 81.6% of cited domains contribute a single URL — and the 18.4% that contribute several are the ones that come back next week
Across 713 domains that ChatGPT, Perplexity and Claude cited while answering 12 small-business buyer questions in four dated weekly runs, 81.6% were cited through exactly one URL; 18.4% through two or more. The instrument is our weekly self-audit: 12 questions × 4 engines × 3 samples = 144 answers per run, on 2026-07-28, 08-03, 08-10 and 08-17. Per engine and run we normalized every returned citation (query, fragment and trailing slash stripped, host lowercased, `www.` dropped) and counted distinct URLs per cited domain; Gemini is excluded as path-blind — its channel returns bare hostnames. The census is a footnote next to the second finding: a domain cited through several URLs in one week was cited again the following week 92.7% of the time (178 of 192); a single-URL domain, 55.0% (482 of 877).
Primary source: AskedAbout — URLs-per-domain cut of four dated four-engine runs (2026-07-28 · 08-03 · 08-10 · 08-17)
What we measured, and how
Four production runs, pulled by runId (2026-07-28 `jh7bvmn1z0tx0s8cwgk7e8xtqn8bd8fw`, 2026-08-03 `jh70zhz8f7xfe62z8vvxysxze18brfvp`, 2026-08-10 `jh75shn14ccqswchppczr9ychh8c7et1`, 2026-08-17 `jh72qsgx56k5490838rt6fa41h8cnwhh`), 576 answers. Each answer's citation array was normalized and deduplicated on the normalized URL: 6,179 citation slots, of which 1,674 — every Gemini entry — are bare hostnames with no scheme or path, so Gemini is excluded from every ratio below as an instrument fact. That leaves 4,505 path-carrying slots from ChatGPT, Perplexity and Claude. Domain = host without `www.`, the production sensor's rule. For each (engine, run) we asked, per cited domain: how many distinct URLs did this engine cite from it across its 36 answers? Pooled figures merge URL sets across runs and engines before counting. Earlier cuts of this corpus found 92.3% of cited URLs are deep pages and most cited slugs name the question; this one asks how many such pages one site gets to contribute.
The census: one URL per domain is the rule
| Engine (pooled, 4 runs) | Cited domains | Single URL | ≥2 URLs | Max URLs from one domain | Mean URLs/domain |
|---|---|---|---|---|---|
| ChatGPT | 379 | 86.0% | 14.0% | 25 (reddit.com) | 1.39 |
| Perplexity | 294 | 82.7% | 17.3% | 30 (reddit.com) | 1.47 |
| Claude | 191 | 82.7% | 17.3% | 20 (trysight.ai) | 1.34 |
| Pooled (3 engines) | 713 | 81.6% | 18.4% | 54 (reddit.com) | 1.53 |
Pooled, 582 of 713 domains contributed a single URL, 131 two or more, 58 (8.1%) three or more; the 713 domains supplied 1,091 distinct URLs. No engine departs from the pattern — single-URL share is 82.7% to 86.0% on all three.
Stable across four weeks
| Run (3 engines pooled) | Cited domains | Single URL | ≥2 URLs | ≥3 URLs | Max (domain) | Mean |
|---|---|---|---|---|---|---|
| 2026-07-28 | 335 | 81.5% | 18.5% | 9.3% | 18 (reddit.com) | 1.41 |
| 2026-08-03 | 348 | 81.9% | 18.1% | 9.2% | 34 (reddit.com) | 1.43 |
| 2026-08-10 | 386 | 82.6% | 17.4% | 9.8% | 23 (reddit.com) | 1.44 |
| 2026-08-17 | 386 | 83.4% | 16.6% | 7.5% | 26 (reddit.com) | 1.41 |
Do multi-page domains survive better?
We already knew that citations which survived one week are far more likely to survive the next. This cut adds a predictor available from a single week: how many URLs the engine cited from the domain. For each of the three week-to-week transitions, per engine, we took every domain cited in week k and asked whether the same engine cited it in week k+1, split by its week-k URL count.
| Engine (3 transitions pooled) | Single-URL domains cited again next week | Multi-URL (≥2) domains cited again next week | All |
|---|---|---|---|
| ChatGPT | 36.2% (145 of 401) | 79.3% (46 of 58) | 41.6% (191 of 459) |
| Perplexity | 68.3% (293 of 429) | 97.8% (90 of 92) | 73.5% (383 of 521) |
| Claude | 58.9% (149 of 253) | 85.0% (34 of 40) | 62.5% (183 of 293) |
| Pooled | 55.0% (482 of 877) | 92.7% (178 of 192) | 61.7% (660 of 1,069) |
The gap holds in all nine engine-transition cells: smallest Claude 08-10→08-17 (61.4% single vs 81.3% multi, n = 83 and 16), largest ChatGPT 08-10→08-17 (26.8% vs 66.7%, n = 138 and 24). Read it as correlation, not causation: a domain cited via several pages is almost by definition one cited more often and on more questions, and higher-citation domains persist for reasons unrelated to page count. What the cut does establish is that a single week's URL count separates the flickering tail from the stable core.
The top multi-URL domains
| Domain | Distinct URLs (4 runs, 3 engines) | Engines | Runs seen |
|---|---|---|---|
| reddit.com | 54 | ChatGPT, Perplexity | 4 of 4 |
| trysight.ai | 31 | Perplexity, Claude | 4 of 4 |
| semrush.com | 12 | ChatGPT, Perplexity, Claude | 4 of 4 |
| help.openai.com | 11 | ChatGPT, Perplexity | 4 of 4 |
| scope.online | 11 | ChatGPT, Perplexity, Claude | 4 of 4 |
| searchengineland.com | 11 | ChatGPT, Perplexity, Claude | 4 of 4 |
| dageno.ai | 10 | ChatGPT, Perplexity, Claude | 4 of 4 |
| linkedin.com | 10 | Perplexity | 4 of 4 |
| openai.com | 10 | ChatGPT, Perplexity | 4 of 4 |
| arxiv.org | 9 | ChatGPT | 4 of 4 |
Our own domain
askedabout.com was cited through one URL per engine, never two: on 2026-08-10 Perplexity cited askedabout.com/compare/best-profound-alternatives; on 2026-08-17 Perplexity cited the same URL again and ChatGPT cited the homepage, askedabout.com. Nothing from Claude or Gemini, nothing in the first two runs. By this cut's reading we are a single-URL domain on each engine — the bucket that comes back 55.0% of the time.
Limits
- Gemini is excluded: all 1,674 of its citation entries are bare hostnames, so a distinct-URL count would measure the channel, not the engine.
- Engines were asked through their APIs, not the consumer apps.
- 12 questions in one category (AI-visibility tooling for small businesses).
- The distinct-URL count is a floor: ChatGPT's frequent homepage citations collapse to one URL per domain, and stripping utm parameters merges tracked variants of one page.
- Survival rates are per-domain, unweighted; multi-URL buckets are small (n = 40 to 92 per engine).
Why this matters if you care about AI visibility
- One page is the normal ticket. 81.6% of cited domains get in through a single URL; the mean is 1.53. You do not need a content library to be cited — you need one page that answers the question, in the shape engines already cite.
- Several pages is the stable ticket. Domains cited via two or more URLs came back the next week 92.7% of the time against 55.0% for one-URL domains. Whether page count causes the stability or merely marks it, a second answer page is the visible difference between core and tail.
- Which bucket are you in this week? With 35 to 93 domains competing per question, most sites that appear at all appear once and vanish. Find out whether, and through which page, you are cited: check which engines name you.
Frequently asked
Does ChatGPT cite multiple pages from the same website in one answer set?
Rarely. Pooled across four weekly runs of 12 buyer questions, ChatGPT cited 379 domains, 86.0% through exactly one URL; reddit.com was the maximum at 25 distinct URLs. Perplexity and Claude both read 82.7% single-URL.
Do websites with more cited pages keep getting cited by AI?
In our data yes, as correlation: domains cited via two or more URLs in one week were cited again by the same engine the next week 92.7% of the time (178 of 192), versus 55.0% for single-URL domains (482 of 877); the gap held in all nine engine-by-week cells.
How many pages does a small business need to be cited by AI search?
One is the norm — 81.6% of cited domains contribute a single URL; mean 1.53. But the one-URL bucket is the unstable one; domains that persist week to week are disproportionately those with two or more cited pages. Full tables: growth/citations/urls-per-domain-cut-2026-08-18.json.
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.