SUN, SEPTEMBER 06, 2026
Independent · In‑Depth · Practitioner‑Tested
✎ General

Two Instruments With No Shared Failure Mode, Pointing the Same Way

A follow-up to our 36-day crawler study, using a second and unrelated measurement. Server logs count requests arriving; third-party citation tracking samples responses being produced. Both show the same ChatGPT decline over the same window, both show Perplexity as the only platform growing — and two further weeks of data correct our earlier framing, because the decline we described as ongoing had already stopped.

By AIToolsRecap September 6, 2026 9 min read 21 views
Home Articles General We Measured the AI Crawler Decline Twice, With ...
METHOD

Two independent measurements over the same period. First-party server logs, crawler traffic identified by user agent. Third-party citation tracking via Ahrefs Brand Radar, which samples AI platform responses rather than reading our logs.

● Different methodologies. Neither can see what the other sees.

● Reported as index values and ratios, baselined on the week of 1 to 7 August.

● One site, one vertical. Direction is transferable, magnitude is not.

The correction first

WE DESCRIBED AN ONGOING DECLINE. IT HAD ALREADY STOPPED

Our 36-day study reported ChatGPT crawl volume falling across three consecutive weeks and did not say whether it had bottomed out.

Two further weeks of data show it had. The decline was a step change that plateaued around 21 August, and the level has held within a narrow band since. That is a cleaner finding than a continuing slide, and it changes what publishers should do about it.

What both instruments show

Week ChatGPT crawl index Bing crawl index
1 to 7 Aug (baseline) 100 100
8 to 14 Aug 79 87
15 to 21 Aug 57 80
22 to 28 Aug 31 80
29 Aug to 4 Sept 33 — flat 86

Two weeks at 31 and 33 is a plateau, not a trend. Daily figures across the most recent week vary by less than 15 percent around that level.

The independent measurement points the same way. Over a seven-day window, third-party tracking recorded ChatGPT responses citing this domain falling by roughly a quarter, with the number of distinct pages cited falling by a third. Copilot fell by a similar proportion.

WHY AGREEMENT MATTERS MORE THAN EITHER NUMBER

Server logs count requests arriving. Citation tracking samples responses being produced. Neither instrument can see what the other sees, and they have no shared failure mode.

A logging artefact would not show up in citation sampling. A sampling artefact would not show up in logs. Both moving together over the same window is the part worth reporting.

The finding nobody is covering

Across the platforms tracked, Perplexity was the only one growing. Over the same seven days it increased both responses citing this domain and the number of distinct pages cited — the page count rising by roughly a third while ChatGPT's fell by a third.

Our server logs corroborate that independently. Perplexity crawl volume has held steady throughout the period in which ChatGPT fell by two thirds, and Claude crawling more than doubled from its August baseline.

Platform Direction over the period
ChatGPTDown about two thirds, then flat
CopilotDown, similar proportion
PerplexityUp on both responses and pages cited
ClaudeCrawl volume more than doubled
Google AI ModeBroadly flat
Google AI OverviewsSmall and declining from a small base

The ratio that surprised us most

Setting the AI platforms aside, one figure from the logs is worth stating on its own: Bing crawls this site at roughly thirteen times the rate Google does.

That ratio has been stable throughout the period. It is not a consequence of the ChatGPT change, and it is not something Search Console can show you — Google reports on Google.

For a publisher deciding where to spend optimisation effort, a thirteen-to-one crawl ratio is a more useful number than any citation count.

What this means for publishers

Finding What to do
Step change, not slideRecalibrate to the new level rather than bracing for further decline
One channel moved, others did notTrack every crawler as its own index. The others are your control group
Perplexity growingNobody optimises for it. Worth knowing which of your pages it cites
Bing at thirteen times GoogleCheck Bing Webmaster Tools. It may be your primary channel and you would not know
Two instruments agreedMeasure twice before publishing a conclusion. One source is an observation

Limitations

Stated plainly, because a study that hides these is not a study.

  • One site, one vertical. An AI tools publication is not representative of news, e-commerce or documentation.
  • Citation tracking is a sample, not a census. Ahrefs Brand Radar estimates from sampled responses. Two instruments agreeing is meaningful; neither is ground truth.
  • User agents can be spoofed. We report what identifies itself.
  • Correlation, not causation. The inflection falls in the same week as a documented retrieval change. That is suggestive and it is not proof.
  • Crawl volume is not citation volume, and neither is referral traffic. Three related measurements, three different things.
  • Two weeks is short for calling a plateau. We will publish an update if the level moves again.

How to run this on your own site

  • Log crawler requests by user agent and group the AI platforms separately. Search Console will not show you this.
  • Track each as an index against a baseline week rather than as raw counts. Relative movement is what tells you something changed.
  • Watch the others as a control. One crawler moving is a vendor change. All of them moving is your problem.
  • Get a second instrument. Citation tracking and server logs have no shared failure mode, which is what makes agreement meaningful.
  • Date every figure you quote. A number that can fall two thirds in three weeks is a snapshot.

Sources

FAQ

What did this study measure?

AI crawler behaviour on one publication, measured two ways — first-party server logs identified by user agent, and third-party citation tracking that samples AI platform responses. Reported as index values against the week of 1 to 7 August.

Why does using two instruments matter?

They have no shared failure mode. Server logs count requests arriving; citation tracking samples responses being produced. Neither can see what the other sees, so agreement between them is more informative than either alone.

Has the ChatGPT decline stopped?

It appears to have. The level has held within a narrow band for two weeks after falling by roughly two thirds. Two weeks is short for calling a plateau, and we will publish an update if it moves.

Which platform is growing?

Perplexity, on both measurements — more responses citing the domain and more distinct pages cited. Claude crawl volume has also more than doubled from its August baseline.

What is the Bing to Google crawl ratio?

Roughly thirteen to one on this site, stable throughout the period. It is not affected by the ChatGPT change and Search Console cannot show it to you.

Can I replicate this?

Yes. Log crawler requests by user agent, index each platform against a baseline week, and watch the others as a control group. A second measurement source is what turns an observation into a finding.

Why report indexes rather than raw counts?

Because the shape is the finding. Absolute volume varies enormously by site and would tell a reader less about whether the same thing is happening to them.

Tags
AI ResearchSEOGEOChatGPTPerplexityClaudeBingCrawlersPublishingData Study2026

Spot an inaccuracy?

We verify facts before publishing and correct errors promptly. If something in this article is wrong or outdated, let us know.

Report an error →
💡 General AI Tools prompts
Prompt Guide
Best Claude AI Prompts for SEO (2026) — Content, Technical, and Comparison SEO
Claude Sonnet 5 and Opus 5 are strong for SEO work that requires writing quality, structured analysis, and long-form content generation. With 1M context, Claude can analyse an entire site's content structure, compare competing pages, and write complete article drafts in one session. These prompts cover the full SEO workflow: keyword research synthesis, content briefs, on-page optimisation, meta descriptions, technical audit interpretation, and comparison content that ranks above AI Overviews.
Get Prompts →
Prompt Guide
Best ChatGPT Prompts for SEO (2026) — GPT-5.6 and Browse
ChatGPT with GPT-5.6 Sol and Browse enabled is a capable SEO research tool — it can search the live web, analyse SERP results, and synthesise content briefs in a single session. GPT-5.6 Terra at $2.50/M offers a cost-efficient option for high-volume SEO content generation. These prompts are optimised for ChatGPT Plus with Browse, the ChatGPT Work product for larger projects, and the OpenAI API with web_search tool enabled.
Get Prompts →
Prompt Guide
Best Claude Opus 5 and Sonnet 5 Prompts for Writing (2026)
Claude Opus 5 and Sonnet 5 consistently produce the highest-quality long-form writing of any AI model in July 2026 — a lead documented across writing benchmarks and user testing since Claude 3 Opus. With 1M context and 128K output on Opus 5, Claude can write book chapters, complete reports, and long-form content without truncating. Sonnet 5 at $2/$10/M (intro through August 31) is the best value writing model available. These prompts are optimised for claude.ai Pro/Max, Claude Cowork, and the API.
Get Prompts →