How long does AI take to notice a website change?

The detection crawl is the first re-crawl of the page that saw the new content. We count days from that crawl. Some answers showed the new amount before our crawler saw it, because the page changed somewhere between two crawls.

By the detection crawl, 12.3% of quoted-old cases already showed the new amount. That rose to 16.1% by day 7, 22.8% by day 30 and 26.5% by day 60. After day 30 the line is almost flat. If an answer has not picked up a new number in a month, it rarely does in the second month.

Among the 187 quoted-old cases where the new amount did appear within 60 days, the median lag was 2 days after the detection crawl. A quarter took 17.5 days or more, and 10% took 36 days or more.

These lags start when our crawler detected the change. The page may have changed earlier, so the time since publication can be longer by up to one crawl interval. The median interval was 15.1 days.

Missing the new number is half of the problem. The other half is repeating the old one. In 48.3% of the 706 quoted-old cases, at least one answer after the detection crawl still quoted a dollar amount that the page had removed.

Among the 341 cases that did, the median last answer with the old amount came 43 days after the detection crawl, and 10% came 58 days or later. ChatGPT page update lag shows up in the old number too: on ChatGPT Search, 73.0% of 37 cases still quoted the old amount after the detection crawl, against 46.9% of 669 cases on Google AI Mode.

Dollar amounts repeat across pages, so some matches happen by chance. To measure chance, we took each case and searched its answers for a dollar amount that a different page had added. That unrelated amount appeared within 60 days in 7.6% of 655 cases. The real new amount appeared in 26.5%, which is 3.5 times as often. The pickup is real, but it is small.

Under a stricter rule, where the answer that quoted the old amount also cited the changed page, the pickup rate was 30.3% of 330 cases. Only 24.6% of the 187 answers that first showed the new amount cited the changed page. The other 75.4% likely took the number from another source that also updated.

With a 30-day window, which lets us include changes detected through September 6, 2026, ChatGPT Search showed the new amount in 36.5% of 126 quoted-old cases. Google AI Mode showed it in 19.5% of 1,603 cases.

Within ChatGPT Search, cases detected in the gpt-5-6 period (from August 8, 2026) picked up the new amount in 39.3% of 89 cases within 30 days. The gpt-5-5 period had only 6 cases, too few to compare. Google AI Mode answered the same dates at 17.3% of 934 cases.

We grouped pages by the type of site they sit on, keeping groups with at least 30 quoted-old cases. Directory sites had the highest pickup at 33.6% of 125 cases, followed by news sites at 32.1% and general sites at 31.8%. E-commerce sites had the lowest at 8.5% of 118 cases.

Most e-commerce cases are product listings whose prices move often. On amazon.com, the new amount appeared in 6.1% of 82 cases. On openvc.app it appeared in 33.3% of 111 cases, and on pcmag.com in 43.9% of 41 cases.

A substantive change is one where at least 20 words of text changed, after we remove counters, dates, links and navigation lists. Comparing the first crawl of a cited page with a crawl about 30 days later, 20.6% of pages had a substantive change. About 90 days later, 33.6% had one.

Review pages changed most: 42.2% within 30 days and 51.9% within 90. Homepages followed at 29.0% and 42.6%. Documentation pages changed least, at 15.0% and 29.7%. Reddit threads changed in 79.5% of cases within 30 days, but mostly in vote counts; only 15.1% had a substantive change.

Change detection needs at least two crawls of the same page. Of 2,244,719 cited pages crawled successfully between March 10 and September 23, 2026, 1,130,889 (50.4%) were crawled at least twice. Those pages produced 3,226,769 back-to-back crawl pairs. The median gap between two crawls of a page was 15.1 days; a quarter were 10.0 days or less and a quarter were 21.8 days or more.

A raw fingerprint of page text changed between 15.5% of crawls taken less than a day apart, which is noise from counters, dates and rotating links. We remove that noise before counting a change. On a sample of 40,329 raw changes, 10.8% were pure noise after cleaning.

For each changed page, we took the prompts and engines that cited it in the 30 days before, and checked their answers in the 30 days after the detection crawl. After a substantive change, 33.4% of 24,521 prompt and engine pairs still cited the same page. Pages that did not change kept 32.7%, after matching the time between crawls.

The difference sits in the rest. After a change, 21.5% moved to another page on the same site, against 16.7% for unchanged pages. So whether AI answers reflect website changes is not visible in lost citations. It shows as a small shift to other pages on the same site.

We compared the set of brands named in the last answer before a change with the first answer after the detection crawl, on ChatGPT Search and Google AI Mode from June 2026. We used overlap: brands named in both answers divided by brands named in either.

After a substantive page change, the overlap was 0.292 across 11,823 pairs. After a crawl where nothing changed, it was 0.298 across 13,092 pairs. For crawls 7 to 21 days apart, the two were 0.292 and 0.281. A change to one cited page did not move the brand list beyond normal day-to-day churn.

We took pages that AI engines cited in the 30 days before they started returning 404 or 410, and stayed that way on every later crawl. On ChatGPT Search and Google AI Mode, for breaks detected between May 29 and August 7, 2026, 13.6% of 557 page and engine pairs cited the dead page again in the next 60 days. Pages that stayed live were cited again in 62.3% of 2,020 pairs.

Most of the leftover citations stopped fast. The median last citation of a dead page came 7.5 days after the crawl that found it broken, and 90% came within 26 days. Only 0.9% were still cited between day 31 and day 60. Our study of how often AI citations break measures how many cited pages are dead; this adds how long the citations last after.

Pricing pages are pages whose address contains pricing, plans or price. Within 30 days, 25.4% of 138 pricing pages had a substantive change, and within 90 days 44.2% of 215 did. Among raw changes to pricing pages, 13.3% of 1,046 replaced a dollar amount. For other pages the rate was 6.6% of 39,283.

Few of those changes had an earlier answer quoting the old price on the same prompt. Only one pricing page case had a full 60-day window, so we cannot time pricing pages on their own. The quoted-old cases above come mostly from review, comparison and product pages that state prices.

We measure from the crawl that detected the change. The page may have changed up to one crawl interval earlier; the median interval here was 15.1 days. We only use crawls from March 10 to September 23, 2026, because our crawler changed how it extracts page text on September 24, which would look like a change to every page. Raw text fingerprints are cleaned with the same rules our crawler now uses, and changes of fewer than 20 words count as small changes, not substantive ones. Some substantive changes are still rotating lists we could not remove.

The dollar-amount cases use every change detected on ChatGPT Search and Google AI Mode citations from May 24 to September 23, 2026, and answers through October 6, 2026. The 60-day figures use changes detected by August 7. ChatGPT and Google AI Overviews stopped running in April 2026, so they appear only in the citation results. We first tried to measure any change in what AI said about a brand, but daily answers change too much on their own for that test to show a lag; brand overlap is reported above instead. We computed the 26.5% headline two ways: a token match in our analysis code and a plain text search of answers in the database, which gave 25.8%.

Do not expect AI answers to repeat a new price or number soon after you publish it. In this data, about 1 in 4 cases picked up a changed dollar amount within 60 days, and about half kept quoting the old one.

Change the number everywhere it is cited, not only on your own page: 75.4% of the answers that showed a new amount did not cite the changed page. Check review, directory and comparison pages that state your prices. Measure from the day you publish, allow at least 30 days, and judge by whether the old number disappears, not only whether the new one shows up. If you retire a page, AI engines stopped citing most dead pages within a week or two, so redirect it to the page that should take its citations.