Does AI mention you at all?
Enter your domain and email. As soon as you confirm the link in your inbox, we start the analysis. The result, with a score from 0 to 100, will appear on your page.
Search ranks you. A model either mentions you or it does not.
A classic search result is a list: ten links, you are one of them, and the visitor picks. A language model answer is not a list, it is a verdict. The model names two or three solutions and leaves the rest out. There is no second page of results to climb onto with a better title.
- Measures your place in a list of links
- The visitor sees ten options
- The click is yours, so is the choice
- A title change moves the ranking
- Measures whether you are in the answer itself
- The user sees two or three names
- If you are not among them, you are nowhere
- Machine readable content decides
The score has two halves and neither is an opinion
Technical readiness is measured on your site, presence inside the model answers. The weights are stored as version 1 and travel with every analysis, so two measurements of the same domain six months apart stay comparable.
Share of mentions is the single largest weight, because it is the only one that directly says whether the model utters your name at all. All technical components together carry less than that one.
Five technical checks, 40 points
All of this is read from your domain and none of it requires access to your systems. Every check reports its own number, not just a yes or a no.
robots.txt and the five AI bots
We read your robots.txt and check how many of the five model crawlers are allowed in. Each blocked one removes a proportional share of the weight: with two of five closed, 12 times 3/5 remains.
Content without JavaScript
We fetch the page the way a machine gets it: without running scripts. Full points mean at least 1,200 characters of text. Below 200 characters with a single page application signature the page is an empty shell and scores zero, because a model does not run your JavaScript.
JSON-LD that actually says something
A valid block earns half the weight, a relevant type the other half. WebSite and BreadcrumbList are deliberately off the list: almost every page has them and they say nothing about the company.
llms.txt
A short file at the root of the domain that tells a model, in readable form, who you are, what you offer and where to go for detail. We check that it exists and that it parses.
sitemap.xml
We check presence, validity and the number of URLs. The lowest weight in the group, because a sitemap alone convinces no model, yet without one half of a site stays undiscovered.
Every request goes through a filter
Because the visitor picks the target, every resolved address is checked before the fetch and again on each redirect. Only http and https, only ports 80 and 443, at most five redirects and at most 2 MB transferred.
Presence in answers, 60 points
The queries are not about your name. We ask the way a buyer asks who does not know you yet, and then read who the model lists.
Share of mentions
How many of the 24 answers name you at all. The only component that says whether the model knows you, which is why it carries the most.
Model coverage
How many of the three providers mention you at least once. A mention in one model only is narrow visibility: a buyer on another model never meets you.
Average position
Where in the answer you appear. The first name counts for more than the ninth; at position ten the component falls to zero, that being the length of a typical recommendation list.
This is what the report looks like
Blank on purpose: we put no measurement on this page as an example, neither someone else's nor our own. Someone else's result is their data, and ours would be advertising rather than a measure. Every row below is filled with your value, measured on your domain on the day you request the analysis.
Every number in the report carries its origin: which check returned it, which model answered and when. None of it is a judgement call.
What happens after you submit
Six steps. The second one matters: until you confirm your e-mail, no analysis runs and no model is called.
The domain is canonicalised through the public suffix list, so example.com and www.example.com are one domain and example.co.uk is not trimmed wrongly.
A signed single use link, valid 24 hours. Until the click there is not a single model call, so a fake sign up costs nobody anything.
robots.txt, llms.txt, sitemap.xml, a fetch without JavaScript and a JSON-LD parse. All through the address filter, with size and time limits.
The three providers run in parallel rather than one after another. The queries derive from your industry and from the languages the site itself declares.
Eight components by the version 1 weights. The version is written into the analysis, so an old score stays readable after the weights are recalibrated.
A signed link to the report, valid 12 months. It contains the score, the breakdown by component and the measured facts of every check.
Four things you can fix this week
The technical 40 points are the easiest part of the score, because they do not depend on what the models think of you. They depend only on what the models find.
1. Open robots.txt to the five bots
Many sites block them without anyone deciding to: default firewall or plugin rules sweep them up together with aggressive scrapers.
User-agent: GPTBot Allow: / User-agent: ClaudeBot Allow: / User-agent: PerplexityBot Allow: / User-agent: Google-Extended Allow: / User-agent: CCBot Allow: /
2. Add llms.txt
One file at the root of the domain, no styling and no scripts. Say who you are, who you sell to and where the detail lives.
# Company name > One sentence: what you do and for whom. ## Solutions - [Solution name](https://...): what it solves ## FAQ - [Question](https://...)
3. Serve the text without JavaScript
Open the page with JavaScript turned off and see how much text is left. If an empty shell remains, a model sees the same emptiness. Server rendering or a static build removes the problem and is not a matter of taste but of reachability.
4. Mark up what you are, not just that you exist
Add JSON-LD with a type that says something about the company. Organization or LocalBusiness for the company, Service or SoftwareApplication for the offering, FAQPage for questions.
Free, with limits we state up front
Every free analysis spends 24 real calls to three models. The limits are not an upsell trick but the reason the service still exists tomorrow.
Free analysis limits
- One domain
- 1 analysis / 30 days
- One IP address
- 3 domains / day
- One e-mail address
- 3 messages / day
- One e-mail address
- 5 distinct domains / 30 days
- Confirmation link
- 24 hours, single use
- Report link
- 12 months
What happens to your data
- The IP address is not stored raw. A hash salted with an application secret is stored, because a plain SHA-256 over an IPv4 address hides nothing: that space is tabulated in minutes.
- Consent to processing is separate from marketing consent. The first is required for the analysis, the second is optional and does not gate it.
- We analyse the publicly available content of your domain. No access to your systems, no credentials, no code on your site.
Frequently asked questions
What is AI visibility?
AI visibility is the question of whether a language model mentions you when someone asks for the kind of solution you offer. Classic SEO measures your place in a list of links; AI visibility measures whether you are in the answer itself. These are two different measurements: a page can rank first in search and still never be mentioned, because its content is only reachable through JavaScript or because robots.txt closes it to model crawlers.
How is the score from 0 to 100 built?
The score has two halves. Technical readiness carries 40 points and is measured on your site: robots.txt and the five AI bots 12, content without JavaScript 10, JSON-LD 8, llms.txt 6, sitemap.xml 4. Presence in answers carries 60 points and is measured inside the models: share of mentions 40, model coverage 12, average position in the answer 8. The weights are stored in the service configuration as version 1 and are never changed silently, because that would make comparison over time a lie.
Which models are actually asked?
Three: Anthropic (claude-haiku-4-5), OpenAI (gpt-4o-mini) and Google (gemini-3.5-flash-lite). The free analysis puts 8 queries to each of the three, so 24 calls, and all three run in one parallel batch. The models live in configuration rather than in code, so the list can change without touching the scoring logic.
Why do I have to confirm my e-mail?
Until you click the link in the e-mail, the service calls no model at all. That is the brake on both cost and abuse: no confirmation means no calls and therefore no bill. The confirmation link is signed, single use and valid for 24 hours.
Where and how long do you keep my data?
The IP address is never stored raw, only as a hash salted with an application secret, so the address cannot be reconstructed from the record. Consent to processing is mandatory and separate from the optional marketing consent. The report link is signed and valid for 12 months.
How often can I check the same domain?
The free analysis is available once per registrable domain every 30 days. On top of that: three domains per day from one IP address, three messages per day to one e-mail address, and five distinct domains per e-mail address in 30 days. The limits exist because every free analysis spends real calls to three models.
See what the models say about you
One domain, one e-mail address, 24 queries to three models and a score from 0 to 100 with every point explained. No card.