AI assistants fetch pages live while they compose an answer, and they don't wait long. WriteWorks times the server response on every page it crawls, once as its own crawler and once with an AI crawler's user agent, and measures the size of the HTML each page sends. Slow responses and pages over the 2 MB that several AI crawlers read are flagged, page by page.
7-day trial · Cancel anytime


This is the working product, not a mockup: the same visibility scores, briefs and published content your team would be looking at. Start a free trial and point it at your own domain, or take the tour of what sits behind the screen.
Live fetches, bot user agents and pages that are simply too big.
When an assistant fetches sources mid-answer, a page that takes two seconds to start responding is easy to drop in favour of one that answers in half a second.
Lab tools measure how a page paints for a person. They don't tell you how quickly your server answers a request with GPTBot or ClaudeBot in the user agent, which can be very different behind a CDN or firewall.
Inline scripts, SVGs and JSON state can push a page past 2 MB. Several AI crawlers stop reading at about that point, and the main content is often below it.
A TTFB checker that tests a single URL won't find the 30 product pages that time out because they skip the cache.
Time it as a bot, apply the thresholds, check the size, track it.
Every page is fetched by the WriteWorks crawler and probed once with an AI crawler's user agent, rotating through GPTBot, OAI-SearchBot, PerplexityBot and ClaudeBot across the crawl. The slower of the two timings is the one that counts.
Under 0.8 seconds to start responding is the target. Between 0.8 and 1.8 seconds is a notice that costs 8 points of bot access; over 1.8 seconds is a warning that costs 20.
The uncompressed HTML of each page is measured. Over 2 MB is an error that costs 30 points of bot access, with the fix: move inline scripts, styles, SVGs and data payloads out, and paginate very long lists.
The Pages view shows TTFB and HTML size for every page and sorts by slowest TTFB. Each page's detail charts its response time and HTML size across crawls, so you can see a fix land.


What each brand fixed, measured before and after with the same audit, in their own words.
“Turning caching on made our pages 2.5 times faster for crawlers. While we were in there, we fixed the headings too: guides with a clean heading structure went from 32% to 100%.”
Two of the bot access checks in AI discoverability.
Time to first byte for every crawled page.
Timed as GPTBot, OAI-SearchBot, PerplexityBot or ClaudeBot.
Uncompressed HTML size, flagged over 2 MB.
Find the slowest pages across the whole crawl in one click.
Pages that error for the AI user agent but load for the crawler are reported.
Response time and HTML size charted crawl by crawl.
From one-off URL tests to every page, every crawl.
Every speed and size check in the crawl.
The people who own the servers and templates.
Server response time and page size across the whole site, as AI crawlers meet them.
The slow templates and oversized payloads, with the fix for each.
Large catalogue and listing pages that drift over the size limit.
A quick read on whether a client's hosting is costing them AI answers.
“We're a booking marketplace, so listings are everything. Since the fixes, pages fully readable without JavaScript went from 84% to 98%, our sitemap dates finally mean something, and our full llms.txt file is real. Over a third of our audited category pages now list their businesses as structured data, with the remaining markets up next.”
Fhresh team
Read the full storyEverything teams typically ask before getting started.
A notice, costing 8 points of bot access. Within crawl limits, but faster sources are preferred.
A warning, costing 20 points. Assistants fetching live tend to drop slow sources.
An error, costing 30 points. Content past the limit is never read.
Pages that load for the crawler but error for the AI user agent, often a CDN or firewall rule.
Audit AI bot accessRelated solutions, adjacent use cases, and platform features.
Server response time with an AI user agent and the 2 MB page size check, on every page, inside the AI discoverability audit.