As of November 21, 2026 · logs from 8 sites, November 1–20, 2026
About 80 times a day, Chrome asks my sites for a file I don’t have: /.well-known/traffic-advice. In the logs, it’s a string of 404 errors from “Chrome Privacy Preserving Prefetch Proxy.” If you check your logs and don’t have a traffic-advice file, sooner or later you’ll run into these lines and ask: is something broken on my site?
The short answer: no, the 404 is fine, because it simply means there’s no advice. What turned out to be more interesting is what’s behind this file: 22,270 page fetches that can look like visits from Google in the logs.
What traffic-advice is
Who’s asking: Private Prefetch Proxy, Chrome’s intermediary server. Through it, the browser prefetches a page you’re likely to click, for example in Google search results. Chrome described it in 2022, when it launched in Chrome 103 on Android.
Why the proxy: the destination site doesn’t learn your IP address before you click. The proxy works like a tunnel and doesn’t see the full URL or the page content. Prefetching happens without cookies, and only for sites where you have no cookies or other stored data.
Why the file: with /.well-known/traffic-advice, a site can ask the proxy for fewer of these connections, or none.
When the file is missing: the proxy uses its default behavior. That’s what the proposal explainer says, and Chrome adds that site owners don’t need to configure anything to benefit from prefetching.
Server logs, November 1–20, 2026, 8 sites I run: an SEO blog, a food blog and travel guides in Polish, English and German. For comparison: 24,344 clicks from Google Search (per Search Console) over the same period.
When is a 404 fine?
Before the proxy sets up a connection to a site, it checks whether the site has left it any advice. A missing file means there isn’t any, so the proxy does what it normally does. The file is fetched by the proxy itself, not by the user’s browser.
So there’s no reason to create the file just to make the errors disappear from your logs. The 404 here is an answer, not a failure. The proxy asks because it wants to know how many connections it can afford, not to evaluate your site. If it gets no limits, it assumes there aren’t any.
Prefetch in the logs: 22,270 pages that look like visits from Google
The request for the file is just the beginning. The prefetches themselves come from different IP addresses than the traffic-advice requests. Google publishes a list of these addresses, and Chrome links to it. When I matched that list against my logs, I got 22,270 page fetches in 20 days. They all look the same: HTML only (not a single CSS file, script or image from those addresses), the referrer https://www.google.com/ and a regular Chrome user agent. More than a quarter came from desktops, the rest from phones.
Did anyone actually visit afterward? In the logs it looks like this: after a visit, the page’s other files (CSS, images) are requested from a non-proxy address, with that page as the referrer. So I counted prefetches that were followed, within 5 minutes, by a fetch of the same page’s files from outside the proxy. There were 5,543 such matches, or 25%. That’s the result of my rule, not the exact share of visits. It could be too high, because on a popular recipe someone else entirely might load the photo in the same 5-minute window. It could also be too low, because someone who has visited before already has some of the files in their browser cache.
| Site | Prefetches in logs | Followed by a fetch of the page’s files (within 5 min) | Clicks in Search Console |
|---|---|---|---|
| ms-fox.pl | 12,044 | 2,755 (23%) | 14,366 |
| madera-online.pl | 7,824 | 2,425 (31%) | 8,494 |
| podrez.pl | 1,440 | 231 (16%) | 422 |
| 5 other travel guides | 962 | 132 (14%) | 1,062 |
| Total | 22,270 | 5,543 (25%) | 24,344 |
Prefetches vs. clicks. Prefetches: page requests from addresses on Chrome’s proxy list. Later fetch: a file of the same page (CSS, image) fetched from another address within 5 minutes, which is a match, not a confirmed visit. Clicks: Search Console, Search, same days.
Prefetches and clicks are two different metrics. Not every click is preceded by a prefetch: the proxy doesn’t work for people who already have the site’s cookies, for example, and some people use other browsers or have page preloading turned off. And not every prefetch ends in a click. podrez.pl shows this best: 1,440 prefetches vs. 422 clicks in Search Console.
The dates show that this is search traffic. On November 11, clicks to madera-online.pl in Search Console jumped from 520 to 1,002, and prefetches on the same domain from 383 to 937. The next day, there were 740 clicks and 873 prefetches.
What does this mean for log analysis? If you count visits from Google by referrer, prefetches will inflate them. On my sites there were almost as many of these “visits” as there were clicks in Search Console, and without the list of proxy addresses you can’t tell them apart from real visits. JavaScript analytics tools probably don’t have this problem, because a prefetched page doesn’t run scripts until someone opens it.
When the proxy asks for the file
Over 20 days, the proxy asked for the file itself 1,583 times across all 8 sites: 67 to 90 times a day, from 89 IP addresses belonging to Google, with no referrer. The fewest requests came around 3:00 a.m., the most around 1:00 p.m. and 8:00 p.m. According to the proposal explainer, the proxy fetches the file when a browser asks to connect to the site. It also refreshes the advice periodically.
The number of requests doesn’t grow with traffic. On ms-fox.pl, which gets the most clicks from Google, there were 23 requests a day on average, roughly one an hour. The proposal explainer gives exactly this kind of refresh as an example. A curious detail: on two sites the 404 page sends a no-store header, which under HTTP rules means it shouldn’t be cached at all. The pattern is the same anyway, so the proxy probably has its own minimum caching time. That part is my hypothesis.
When the file makes sense
When you want to limit prefetch traffic, for example because of a weak server, heavy pages or limits on your hosting plan. The file must be served as application/trafficadvice+json and looks like this:
[{
"user_agent": "prefetch-proxy",
"fraction": 0.3
}]
fraction, from 0 to 1, tells the proxy what share of prefetches to let through. Here it’s about 30%. Instead of fraction, you can set "disallow": true, and the proxy won’t set up any connections for prefetching.
Watch the content type. The file has no extension, so a typical host will serve it as plain text or as a binary file, and Chrome will ignore it. On Apache or LiteSpeed, you can set the type in .htaccess:
<Files "traffic-advice">
ForceType application/trafficadvice+json
</Files>
During a sudden traffic spike, you have another option that doesn’t involve the file. Prefetch requests through the proxy carry the Sec-Purpose: prefetch;anonymous-client-ip header. Your server can answer them with a 503 and a Cache-Control: no-store header, and use a Retry-After header to say when to try again.
Check your site
My tool Can AI bots read your site? shows what your server returns at /.well-known/traffic-advice. With a 404, you’ll see that Chrome uses its default behavior. If the file exists, the tool reads the prefetch-proxy entry (fraction or disallow) and warns you if the file is served with a type other than application/trafficadvice+json, because then Chrome ignores it.
Example from November 21, 2026: en.wikipedia.org has a file with "disallow": true, so Chrome doesn’t prefetch its pages through the proxy.
The real cost: a WordPress 404
Back to the file itself. Six of my sites answer requests for /.well-known/ with a lightweight 404 page straight from the server. The other two, podrez.pl and ms-fox.pl, load WordPress every time and render a full error page with the menu, footer and scripts. Why the difference? The .htaccess file on those six contains one line that the Really Simple Security plugin (formerly Really Simple SSL) adds when it sets up a Let’s Encrypt certificate.
| 404 response | HTML size | Transferred (average from logs) | Time to first byte |
|---|---|---|---|
| Full WordPress page (podrez.pl, ms-fox.pl) | 149–170 KB | 38–41 KB | 0.7–0.9 s |
| Lightweight server 404 (the other 6) | 10 KB | 4.3 KB | 0.2 s |
What a single 404 costs. Size and time: three measurements on November 21, 2026. Bytes transferred: average from logs, November 1–20.
On its own, that’s a small thing. But traffic-advice is only one of the visitors. Over 20 days, podrez.pl and ms-fox.pl received 3,063 requests to /.well-known/: requests for app, security and AI agent files, plus scanners looking for vulnerabilities. Of those, 2,587 got the full WordPress 404 page. The rest were redirects, access denials, lightweight 404s and a few URLs handled by a plugin. WP Rocket only caches 200 responses, so 404 pages never make it into the cache, and every one of those requests starts WordPress from scratch.
The Really Simple Security line looks like this:
RewriteRule ^.well-known/(.*)$ - [L]
It tells the server: don’t pass /.well-known/ URLs to WordPress. It has to go above the WordPress rewrite rules block. If a file exists there, the server serves it directly, and a missing file ends in a lightweight 404. Before you add it, check one thing: whether any plugin handles URLs under /.well-known/ through WordPress. On two of my sites one does, so this line would break something there. That’s why I’m not adding it for now.
Who else comes knocking on .well-known
GoogleAssociationService requests /.well-known/assetlinks.json from the same pool of Google addresses that sends the traffic-advice requests. That’s the Digital Asset Links file, which connects a website with an app, for example on Android. In my logs that’s 1,645 requests in 20 days, even more than traffic-advice. If you don’t have an app, there’s no reason to panic here either.
The new /.well-known/ard.json file was requested mostly by Lighthouse during the same period. I explain why, and whether you need it, in ard.json: does your site need it?
What to do about it
- Nothing, if your server can handle it. A 404 for traffic-advice means the usual rules apply. It’s not an error to fix.
- When you count traffic from logs, filter out prefetches. Drop requests from addresses on Chrome’s proxy list, or add the
Sec-Purposeheader to your log format. Otherwise you’ll overcount visits from Google. - Check what your server returns under
/.well-known/. If it’s a full WordPress 404 page, consider a lighter 404, but first make sure no plugin uses those URLs. - Only add the file if you want to limit prefetching. Remember the
application/trafficadvice+jsontype.
If logs are new to you, start with Server log analysis. You can check AI bots in your own logs with the AI bots in logs tool.
Frequently asked questions
What is Chrome Privacy Preserving Prefetch Proxy in my logs?
It’s Chrome’s intermediary server. The browser uses it to prefetch pages, for example from Google search results, without cookies and without revealing the user’s IP address. Before it sets up a connection, it asks for the /.well-known/traffic-advice file.
Is a 404 for /.well-known/traffic-advice an error?
No. According to the proposal explainer, the proxy uses its default behavior on a 404. Chrome also says you don’t need to configure anything to benefit from prefetching.
How do I limit prefetching through Chrome’s proxy?
Add a /.well-known/traffic-advice file served as application/trafficadvice+json with an entry for prefetch-proxy: fraction from 0 to 1 lets a share of the connections through, and "disallow": true turns them off completely. During a sudden traffic spike, you can also answer requests carrying Sec-Purpose: prefetch;anonymous-client-ip with a 503 and a Cache-Control: no-store header.
Does prefetching inflate traffic stats?
In server logs, yes. Across my 8 sites there were 22,270 prefetches with a Google referrer in 20 days. One in four was followed within 5 minutes by someone fetching files of the same page, but the logs can’t tell whether it was the same person. You can recognize prefetches by the addresses on the proxy list Google publishes. JavaScript analytics tools probably don’t count them, because a prefetched page doesn’t run scripts until someone opens it.
Private Prefetch Proxy, Traffic Advice, proposal explainer, accessed November 21, 2026.
Google, Chrome Prefetch Proxy Geofeed, list of proxy addresses, updated March 18, 2025.
MDN, Speculation Rules API, accessed November 21, 2026.
Chrome Help, Speed up Google Chrome, accessed November 21, 2026.
Google, Digital Asset Links, accessed November 21, 2026.
Own data: server logs from 8 sites and Search Console, November 1–20, 2026, 404 response measurements, November 21, 2026.
Want to see my posts more often on Google?
Add podrez.pl as a preferred source on GoogleMore on this topic: AI search & GEO


