Cited Layer › Services › AI traffic check
Service
AI traffic check
An AI traffic check answers one question from your own server logs: what are ChatGPT and Perplexity actually doing on your site — and is the AI traffic in your analytics real?
Since October 2026 ChatGPT passes its tag and referrer on fewer clicks. Dashboards now show collapses that did not happen, and can hide ones that did. Fourteen days of logs settle it. No prompts are run and nothing is installed. Three working days, $400.
Delivered in writing. No call is required at any point.
Why logs
Analytics showed a collapse. The logs showed what was real.
On one store in October 2026, visits tagged as coming from ChatGPT fell from about 23 a day to 2–3. In the same days, visits with no referrer rose from 11 to 26, landing on the page ChatGPT clicks used to land on. Together the two fell by about 15%. The number that no tracking change can hide — requests from ChatGPT's live fetcher in the server log — fell from 26.5 a day to 7.
Analytics alone would have reported a channel lost. Two of three numbers would have got the size wrong; the first version of the published write-up did exactly that and was corrected. The full measurement, with the correction →
What is checked
Five things, all from the logs and the live site
-
Which pages the engines actually read
Requests from ChatGPT-User, OAI-SearchBot, GPTBot, PerplexityBot, Perplexity-User and ClaudeBot, day by day and page by page. Every address is matched against the ranges the engines publish; scanners using their names are removed first — on one domain they were most of the apparent AI traffic. On a young domain the engines read only the three pages external links pointed to. If that is your case, writing pages for questions cannot work until links point at them, and you should know that before paying for content.
-
Which questions already bring the engines to you
When ChatGPT's live fetcher opens a product or brand page, someone has just asked about that product or brand. Those fetches show which question shapes already reach you — on one store, every one of them named a brand or a model, and none came from a general "where to buy" question.
-
Real drop or lost tracking
Three numbers per day, side by side: clicks tagged as ChatGPT, direct visits by landing page, and live fetches. If direct rises by about as much as tagged clicks fall, on the pages ChatGPT used to send people to, it is mostly tracking.
-
What your structured data claims
Dates that regenerate as "today" on every build — one site was announcing a new video upload date on 226 pages every thirty minutes. Numbers in the markup that appear nowhere on the page, such as a return window no buyer can see. Both take a minute to check and are invisible in analytics.
-
Access and errors
robots.txt as each AI agent sees it, the status codes each one received, and whether a spike of 404s is broken links or scanners knocking on the server's default address.
Deliverable
What you receive
- The engine page list — every URL each AI agent took in fourteen days, verified addresses only.
- The question shapes — the products, brands and pages live fetches were opened for.
- The daily table — tagged clicks, direct by landing page, live fetches, and every crawler as a control.
- The markup findings — dates and numbers in your structured data that contradict the page, with the line to change.
- The commands — so your team can repeat every count without me.
Written report, three working days after the logs arrive. English or Ukrainian.
What is needed
Fourteen days of raw access logs
Nginx or Apache access logs, or an export from your CDN (Cloudflare, Fastly, CloudFront) with user agent, address, path, status and referrer. A read-only file is enough; no access to your systems is required.
Hosted platforms that do not expose raw logs cannot be checked this way. If that is your situation I will say so before anything is invoiced, and the category source map is the alternative.
Limits
What the check does not show
Why an engine changed its behaviour
The logs show that the live fetcher came less often, not whether you were picked for fewer answers or the engine changed how it retrieves. That decision is made inside the engine, and the report will not pretend otherwise.
Who the engines name
No prompts are run. A fetch is the engine reading a page, not a buyer arriving, and not a citation. For who is named and which sites decide your category, see the category source map.
Price
Four ways in
| What | Scope | Price |
|---|---|---|
| Category check | Four of your buyers' questions in ChatGPT and Perplexity. The answers verbatim, every source listed under its answer. No scoring. Two working days. | Free |
| AI traffic check | Fourteen days of your server logs: which pages ChatGPT and Perplexity actually read, whether a ChatGPT drop is lost tracking or lost visits, and what your structured data claims that your pages do not. No prompts. Three working days. | $400 |
| Category source map | Ten buyer questions, three runs each, two engines. The two to four third-party sites that decide answers in your category, whether you are on them, what each requires, and which popular tactics do not work in your category. Five working days. | $600 |
| Placement program | Getting onto those sites, fixing what the logs showed, monthly re-measurement and ChatGPT fetcher monitoring. Guaranteed: the work and the measurement. Not a citation. Three months minimum. | from $1,500/mo |
Questions
Frequently asked
What is an AI traffic check?
A reading of fourteen days of your server logs that shows which pages ChatGPT, Perplexity and other AI agents actually request, whether a change in AI traffic in your analytics is real or a tracking effect, and what your structured data claims that your pages do not.
Why did my ChatGPT traffic drop in October 2026?
On the store measured here, mostly because the clicks stopped carrying ChatGPT's tag and referrer and were counted as direct visits instead. Separately, ChatGPT's live fetcher came about four times less often. Your logs show which of the two happened to you, and in what proportion.
Why not just use analytics?
Analytics sees visitors who arrive with a tag or a referrer. It does not see the engines reading your pages, and it cannot tell an untagged ChatGPT click from someone typing your address. The server log records every request regardless.
Do you need access to my systems?
No. A read-only export of the access logs is enough. Nothing is installed.
What if my platform does not give me raw logs?
Then this check cannot be done, and you will be told so before anything is invoiced. The category source map works without logs.
Send fourteen days of logs
Or ask first whether your setup can produce them. The answer comes back the same day, in writing.
alex@citedlayer.com