How we score AI visibility, out of 100
Every score we publish comes from the ten categories below. That includes the 51 to 90 and 54 to 91 results on our examples page, and every number in our study of 101 South African businesses. The categories are equally weighted, ten points each. Nothing is weighted in a way you cannot see here.
1. AI crawler access
Whether AI crawlers are allowed to read your site at all. We request your robots.txt and check it against the current list of AI crawler names, and we check that it does not accidentally block the stylesheets and scripts your own pages are built from.
What earns the points. 10 if nothing is blocked. 5 if some crawlers are blocked. 0 for a blanket block, or for a rule that blocks the files your page needs to render. No robots.txt at all scores 8, because nothing is blocked but nothing is welcomed either.
2. AI guide file
Whether you publish an llms.txt, the plain text file that tells an AI assistant who you are, what you sell and which pages matter, written for machines rather than for search rankings.
What earns the points. 10 if it exists and is well formed, with a clear summary and curated links. 0 if it is absent, which is still the common case. Points come off if it contradicts the rest of your site.
3. Machine-readable business facts
Your structured data. We read every JSON-LD block on every page and check that it parses, uses the right types, and actually describes your business rather than a generic template.
What earns the points. 10 for complete, correct business and FAQ markup. 5 for a partial or plugin-generated minimum. 0 for none. We also deduct for markup describing things a visitor cannot see on the page.
4. Answer-first content
Whether the first screenful of real text on your homepage says who you are, what you do and where you are. Not the hero image, not a slogan. Text an assistant could quote to answer the question, what is this business.
What earns the points. 10 if the opening paragraph answers all three. 0 if the homepage is images and slogans with no factual text underneath them.
5. Question coverage
Whether your site answers the questions your customers actually ask, in question and answer form. Hours, pricing, service areas, turnaround, delivery, whatever fits your trade. AI assistants quote this format more than any other.
What earns the points. 10 for thorough coverage of what a real buyer asks, marked up so a machine can read it. 0 for none, which was true of 59 of the 101 businesses in our study.
6. Essential business facts
Phone, address or service area, opening hours and a price indication, present as text a reader can see, and consistent everywhere they appear.
What earns the points. 10 for all present, visible and consistent. A fact that exists only inside your structured data does not count as present. We have watched two engines report a business as having no named owner while the owner's name sat in the markup the whole time.
7. Site readability for AI
Whether the meaningful content is in the raw HTML or only appears after JavaScript runs, because many AI crawlers do not run JavaScript. We also check for age gates, consent walls and login walls sitting in front of your content, and for hidden text that contradicts what the page displays.
What earns the points. 10 if everything important is in the raw HTML with sensible headings and titles. Heavy deductions for content that only exists after rendering, or for a gate a crawler cannot get past.
8. Independent reputation signals
What the rest of the web says about you. Google Business Profile, reviews on platforms you do not own, directory listings, and mentions from other sites. This is the one category we report on but cannot fix for you.
What earns the points. Scored on what actually exists and can be found. We report it honestly rather than pretending a technical fix covers it. In our own study, businesses with strong independent reputation averaged 54.4 out of 100 against 42.6 for weak ones.
9. Live AI engine test
The part that is not a checklist. We ask real AI assistants the question a stranger would actually type to find a business like yours, and we record what they say back, word for word, with the date and the engine named.
What earns the points. 10 if you are recommended for the broad category question a cold stranger would ask. 0 if you are not mentioned at all. We always use the broadest, most common phrasing and never your own marketing language, because a question built from your own words is an easy test that tells you nothing.
10. Freshness and trust signals
Whether the site looks current and trustworthy. Recent updates, a correct copyright year, HTTPS, a real about page, working links, and privacy and terms pages that actually exist. We also check whether the site contradicts itself or its own social profiles.
What earns the points. 10 for a current, consistent, complete site. Deductions for stale dates, dead links, missing legal pages, and for facts that differ between your site and your own listings.
What the final number means
80 to 100, AI-ready. Rare. Two of the 101 businesses in our study reached it.
60 to 79, visible but underperforming. An assistant can find you and describe you, but will often pick a competitor.
40 to 59, barely visible. The largest band. 59 of the 101 businesses we audited sat here.
0 to 39, invisible to AI. Most small local businesses, before anyone has looked at this.
What this score is not
It is not a Google ranking, and a good score here does not promise one. It is not a prediction either. It measures what an AI assistant can find, read and verify about your business on the day we test it.
The live engine test is the one category that can move on its own, because AI assistants are not deterministic. The same question can get a different answer from the same engine on the same day. That is why we record the answer word for word with the date and the engine named, and why we re-test rather than treat one clean result as permanent.
We score ourselves the same way
An independent diagnostic run against this website on 6 September 2026 scored it 58.5 out of 100 on this exact rubric. That puts us in the same barely visible band as most of the businesses we audit. Our two weakest categories were independent reputation and the live engine test, which are the two hardest and the two that cannot be fixed by editing your own website.
Publishing that is more useful to you than hiding it. It shows the method is applied consistently, that off-site authority is genuinely the hard part, and that a business selling this is not exempt from it.
Want your own score?
The audit is free. Send us your website address and we will run all ten categories, test you live against ChatGPT plus Gemini or Grok, and send back your score with a plain English list of what is holding it back. Get your free audit.