Methodology
How we check, and where our checks stop
The SiteSignal methodology is a fixed set of automated checks for access, clarity, SEO basics and speed, plus a question test in which a language model reads only your website text. It measures readiness, not rankings or live AI answers.
1. Loading your site
We request your public homepage and a small number of supporting files, such as robots.txt. We do not log in, submit forms or crawl your whole site.
2. Speed and page experience
When Google PageSpeed Insights is configured, we request its mobile and desktop results for your homepage: lab metrics such as Largest Contentful Paint (LCP) and Cumulative Layout Shift (CLS), and field data such as Interaction to Next Paint (INP) when Google has enough real-user data for your site. Field data is often missing for small sites; then we say so instead of guessing. If PageSpeed is not available, the speed check is marked as not measured and does not count against you.
3. SEO basics
We look for the signals search engines rely on to describe a page: title, meta description, headings, language and structured business details where present.
4. AI access
We read your robots.txt and report which known AI bots are allowed, blocked or unclear. Allowing a bot is not a guarantee that any assistant will mention you. Some businesses block training bots on purpose, and that can be a fair choice.
5. The question test
A language model receives only the text we extracted from your site and is asked four plain questions: what is the company called, what does it offer, where is it, and how can someone contact it. Empty answers mean your site does not say it clearly enough.
This is not a conversation with ChatGPT or any other consumer assistant. Different models can answer slightly differently, and the same model can vary between runs. If the model is unavailable we mark the test as unavailable instead of scoring it against you.
What raw HTML can and cannot show
We read the HTML your server sends. Content that only appears after JavaScript runs, inside images, or behind logins may be invisible to us, and to some bots. Schema markup is read from that HTML: if it is injected later by scripts, we may miss it.
Financial scenarios: assumptions, not measured losses
When the homepage identifies a business niche and city, we prefill a typical average customer value and monthly customers acquired through website/search, including quote-based local services. The configured AI proposes editable starting values; a niche/region planning table supplies missing numbers. These are uncertain assumptions, not verified local market benchmarks, and their basis and confidence are shown. A known city can establish region and currency; language alone cannot. Any assumed fallback currency is disclosed. Subscriptions use first-month value; services use completed-job value. Missing business/location information still needs manual inputs.
The calculator is independent of the readiness score. A failed available check contributes 1 issue unit and a warning 0.5, capped at 4 units in each category. We apply a hypothetical additional-customer range of 1–2% per unit to the baseline monthly website/search customers: at most 12–24%. These are planning assumptions, not experimentally established conversion rates. The result is multiplied by average customer value; unavailable checks are excluded.
Fixes receive non-overlapping shares of that same scenario. Displayed shares are rounded down and cannot sum above the total. Actual interactions and results can differ. Changing the city, niche, country or currency does not fetch new market data or convert currencies; review the two financial inputs yourself.
The public preview can show the original estimate. After email unlock, sliders and numeric inputs let you edit it. Edits stay in your browser tab, survive reloads and never modify shared audit findings or another visitor's inputs. No estimate is fabricated if a baseline is missing.
How the scores work
Each area gets a score out of 100 from the checks that actually ran. Checks that could not run are marked 'not measured' and excluded rather than scored as failures, and the report says when a result is partial. The overall score combines the areas; the AI readiness score combines bot access and the question test. Exact weights, missing-check handling and current limits are listed below from our live configuration.
Scores are SiteSignal's assessment of your website, not a ranking from Google, ChatGPT or Perplexity, not a measure of live recommendations, and not a count of lost customers.
Questions owners ask
The current provider and model are shown on this page. Reports record the model that interpreted them. Older reports created before we recorded this show 'not recorded'; we do not assume they used today's model.
Server response time varies, your site may have changed, and language models can give slightly different answers to the same text.
No. Some fix suggestions link to partners, and some are affiliate links. Scores are calculated the same way regardless.
Other free tools · same scan engine
- AI visibility checkerCan AI read and describe you?
- Website graderOne grade, three lenses
- Compare your website with a competitorYou vs. them, same checks
- Robots.txt AI bot checkerIs GPTBot or ClaudeBot blocked?
- Schema markup checkerAre your business facts labelled?
- Page speed checkErrors and slow responses