Website Health Check

The free Website Health Check from AmICited (FlowHunt's free AI brand-visibility tracker) runs a single pass over a site you name and reports back in about 40 seconds. This guide explains exactly what that pass looks at: the robots.txt rules that let or block AI assistant crawlers and ordinary search crawlers, the llms.txt file if one exists, the sitemap and how many URLs it lists, five sampled pages checked for load time, page weight and markup problems, and real-world speed data from the Chrome UX Report where Google has it. It also covers how those checks add up to a 100-point score across four pillars, and how to read the ranked list of fixes the tool produces at the end.

Nothing about the check is saved and nothing about it is exhaustive. It samples five pages rather than crawling the whole site, and a site with no Chrome UX Report data is not a site that failed a test, it is a site Google has not measured yet. Treat the score as a snapshot of what five pages and three files say about the site today, not a guarantee about the rest of the site or a forecast of traffic.

What the check measures

The check pulls from five sources: robots.txt, llms.txt if the site has one, the sitemap, five sampled pages, and the Chrome UX Report for the site's origin. Those results roll up into four pillars, each with a fixed number of points available:

Speed, worth up to 40 points: time to first byte and page weight measured on each of the five sampled pages, plus real-world load data from the Chrome UX Report when Google has recorded any for the site.
Crawlability, worth up to 25 points: whether robots.txt is present and parses, whether it allows the AI assistant crawlers, such as GPTBot, ClaudeBot and PerplexityBot, whether it allows ordinary search crawlers, whether a sitemap is declared and parses, and how many of the sampled URLs redirect instead of loading directly.
AI readability, worth up to 20 points: whether llms.txt exists and how it scores against the same validator behind the llms.txt generator, whether the sampled pages carry structured data, whether the title and meta description are present and within length, whether each page has exactly one H1, and whether a canonical tag is declared.
Reliability, worth up to 15 points: whether the sampled pages answer with a normal 200 status, how many fetches error out or time out, whether any page sits behind a redirect chain longer than one hop, and whether an HTTPS page loads mixed content.

Why AI assistants need more than classic SEO

A search engine can index a page today and let ranking signals catch up over months. An AI assistant does not have that luxury. When someone asks it a question, it fetches or retrieves the page in the moment and reads whatever is there. A blocked crawler, a page with no title, or a paragraph too thin to answer the question costs the assistant an answer it could have given, and it costs the site the mention that answer would have carried.

That is why the check treats AI readability as its own pillar rather than folding it into ordinary SEO. It checks robots.txt against the AI assistant crawler list separately from the ordinary search crawler check, because a site can allow Googlebot while blocking the crawlers that would otherwise read a page and cite it in an answer, and that mistake will not show up in a classic SEO audit.

How the 100-point score works

The 100 points split across the four pillars: 40 for speed, 25 for crawlability, 20 for AI readability, 15 for reliability. Each of the 21 checks inside those pillars is worth a fixed number of points, and a check that cannot run, such as the AI-crawler and search-crawler checks when there is no robots.txt, or the speed checks when the Chrome UX Report has no data for the site, is skipped rather than scored as a failure. A skipped check costs nothing.

The score comes with a short, ranked list of fixes. The ranking and the points behind each fix are computed directly from the check results; a language model only writes the explanation, covering why the issue matters, what it is currently costing the site, what to do about it, and how to verify the fix worked.

Reading the results

Each of the five sampled pages, the home page plus a spread across the sections the sitemap lists, comes back with its own time to first byte, page weight and HTTP status, alongside markup problems such as a missing or overlong title, a missing or duplicate H1, images with no alt text, or a page with no outbound internal links. A slow or heavy page here is a real page on the site, not an estimate.

robots.txt, llms.txt and the sitemap tell three different stories. robots.txt says who is allowed to crawl the site at all. llms.txt, when it exists, says which pages a model should treat as the important ones, and it is graded against the same validator as the llms.txt generator. The sitemap says how many URLs the site is telling crawlers about in the first place, which is often the first sign of a section nobody remembered to add. A missing Chrome UX Report is a data gap, not a low score. Google only publishes that data for origins with enough real Chrome traffic, so a new or low-traffic site will legitimately have none, and the check skips those points instead of marking the site down for it.

What to fix first

The ranked fix list exists because not every issue is worth the same number of points, and the cheapest fixes are often the ones with the biggest recoverable score. A few show up often:

A robots.txt rule that blocks an AI assistant crawler by accident, the single highest-value crawlability check at 8 of the pillar's 25 points.
A missing or generic title or meta description, which is quick to write and directly scored.
Images with no alt text, especially on pages that lean on visuals to make their point.
A missing canonical tag, which costs 2 of the AI readability pillar's 20 points and is one of the cheapest fixes on the list.
A sitemap that has gone stale or was never submitted, which is why the sitemap check counts URLs rather than assuming one exists.

What the free check does not do

The free check is a single pass over five sampled pages, run once, with nothing saved afterward. That is enough to catch the kind of mistake described above, but it will not catch a problem confined to the pages outside the sample, and it cannot tell you whether a fix made last month is still holding.

Tracking a domain in AmICited covers the rest: every page rather than a sample, checks that run again on a schedule instead of once, and a history to compare against instead of a single snapshot.

Frequently asked questions

How long does the check take, and is anything saved?

It takes about 40 seconds. The check runs once against the site you name and nothing about it is stored afterward.

Why does it only sample 5 pages instead of the whole site?

Five pages, the home page plus a spread across the sitemap's sections, is enough to catch systemic problems, like page speed or a repeated missing title, without turning a free one-off check into a full crawl. A five-page sample is a sample: it will not surface an issue confined to pages outside that five.

What does it mean if my site has no robots.txt?

It means the two checks that depend on it, whether AI assistant crawlers are allowed and whether ordinary search crawlers are allowed, are skipped rather than scored as failing. A missing file is treated as missing information, not as a block on every crawler.

Does a low score mean my site is broken?

No. The score adds up 21 checks worth a fixed number of points each, and it reflects what those checks found on the pages and files the tool looked at. A low score points at specific, fixable issues on the fix list; it is not a verdict that the site does not work.

Do I need an llms.txt file to score well?

No. llms.txt is read and graded if the site has one, using the same validator as the llms.txt generator, but a site without one is not penalized for a file it never claimed to have. It is one input among 21, not a requirement.

What does it mean when the Chrome UX Report has no data for my site?

It means Google has not recorded enough real-world Chrome traffic for the origin to publish a report, which is common for newer or lower-traffic sites. The check skips those points instead of scoring the site down, since no data is not the same as bad data.