How the AI Visibility Score is measured
Our analyzer, the tool that checks your website, gives it a score out of 100 for how well AI assistants such as ChatGPT can find it, understand it and recommend it. This page shows how that number is worked out: what each part is worth, where we get the facts, and when a problem may cost points. Every number here comes straight from the code that runs the check, so when the scoring changes, this page changes with it.
The short version
- What the score says
- Whether AI assistants name your business when a customer asks them for what you sell, and whether your site gives them what they need to do it. 100 is the best possible result.
- Where it comes from
- We ask four assistants real buying questions and read their answers. We also read your site, see whether it shows up on Google and Bing, look for your Google Maps listing and test whether an AI agent, a program that buys or books things for a person, could place an order or a booking with you.
- What can cost you points
- Only a problem we can show you: the page, the line in one of your website's files or the measured number. If we cannot show it, it is listed for you to review and costs nothing.
- What never costs you points
- Our own failures. If a security service on your site blocked our tool, if a service we rely on went down or if we cannot search Google in your country, that part is left out of the score. It is never counted as a zero.
- Your kind of business
- A plumber is not asked for a pricing page, and a law firm is not asked for an online checkout. We first work out what kind of business you are and judge you only on what fits it.
- Your own decisions
- If something we flag is deliberate, you can mark it as intentional. For example, a shop may tell crawlers, the robots search engines and AI companies send to read websites, to stay out of its customers' account pages. Once marked, it stops coming back as a new problem.
Ten parts, one hundred points
The score is made of ten parts, and they are not worth the same. The biggest one is whether the assistants name you at all. Two parts apply only to some businesses. Google Maps counts only if customers find you on a map. Agent Access, whether an AI agent can act on your site, applies to software, shops and venues, and even there it counts only when it raises your score, so the chart draws it as an outline with no fill.
100
points when every part is measured
- 16AI Share of Voice
- 13Retrieval Readiness
- 11Search Presence
- 12AI Identity & Crawlability
- 10Technical & Performance
- 6AI Sentiment
- 6Mention Position
- 6Competitor Gap
- 10Local Presence (Google Maps)only for a business found on a map
- 10Agent Accesscounts only when it raises the score
What each part checks, and what happens when we cannot check it
Points are relative. If a part could not be checked, the score is worked out from the parts that were, and the missing part is shown with its reason instead of a silent zero.
16AI Share of Voice
What it checksHow often the assistants name you: the share of questions whose answers named you, counted per question, not per answer. Competitors are counted by the same rule.
When we cannot check itEvery audit asks the assistants, so if this part is missing, something failed on our side.
13Retrieval Readiness
What it checksWhether your site has the pages an assistant needs before it can quote you, judged by your kind of business, and whether your homepage says plainly what you do, for whom and how you differ.
When we cannot check itIf our page reader fails and can only guess from page addresses, a page it did not find proves nothing. The part is then left out instead of charging you for our failure.
11Search Presence
What it checksWhere your site appears on Google for the same buying questions, leaving out the ones that contain your name, and whether Bing has your site at all.
When we cannot check itGoogle and Bing are scored separately, and whichever did not answer is left out. If we cannot search Google in your country, for example in Russia or Belarus, the part is left out instead of searching the US.
12AI Identity & Crawlability
What it checksWhether the programs assistants send to read websites (crawlers) are allowed in, and whether your pages carry the labels in their code that tell an assistant what your business is. We read your files at the site's root, whatever address you typed.
When we cannot check itIf your robots.txt or sitemap refused us or did not answer, we mark it as unknown, not as missing: its points are left out and no fix is raised.
10Technical & Performance
What it checksGoogle's PageSpeed test of your site: speed, basic search setup, accessibility and good practice.
When we cannot check itIf Google's test fails, that says nothing about how fast your site is. The part is left out instead of lowering the score.
6AI Sentiment
What it checksHow the answers describe you when they do name you, judged by a different AI model from the one that wrote the answer.
When we cannot check itIf you were never named, there is nothing to judge. The first part already takes the full cost of that.
6Mention Position
What it checksWhere you appear among the businesses named in an answer. First place earns twice the points of second and three times those of third.
When we cannot check itIf you were never named, you have no position. That is the same fact as not being named, and it is already counted there.
6Competitor Gap
What it checksOf the answers that named any business, the share that named you. An answer is lost when it names a competitor and leaves you out; one that names you is kept, however many others it lists.
When we cannot check itIf no competitor was named either, there is nothing to compare. An answer that names nobody is not proof that you have no competitors.
10Local Presence (Google Maps)
What it checksFor a business customers find in a place: whether your Google Maps listing links to your site or shows your phone number, whether you have claimed it and listed opening hours, where you appear when a customer searches for your kind of business in your city, and whether you have fewer reviews or a lower rating than the listings above you.
When we cannot check itIt does not apply to a business customers do not look for on a map, such as software, an online shop or a publication. A firm, a manufacturer or a business whose kind we could not tell is scored only if we find a listing. A listing with your name that nothing on your site confirms, or a country where we cannot search Maps, is not scored either.
10Agent Access
What it checksWhether an AI agent can act for a customer, not just read about you. For software: your own MCP server and how good it is by the rules of our Atlas catalogue. For a shop: the door your platform provides (Shopify, Tiendanube or Wix) and whether an agent can find a product, fill a cart and check out. For a venue: booking through a partner that Google's AI Mode books with. Half the points are for having a door, half for how well it works.
When we cannot check itIt applies to software, online shops and venues. For any other business it is left out. A door that asks the agent to sign in counts as present, with its quality not yet proven. If we ran out of time looking for a door and found none, the part is left out as our failure. Reports made before 26.10.2026 show it as not measured.
Total, when every part was measured: 100
A problem costs points only when we can show it
Before anything we find can lower your score, it has to pass two tests: our own check worked, and we can show you the proof, such as the page, a line in one of your website's files or the measured number. Say a bakery's report claimed it had no sitemap, when in fact its security service had blocked our tool. This rule exists to stop that, and it has covered every part of the report since 31.10.2026.
Something we found
Did our check work?
Our failure
That part is left out of the score and marked as not checked. It is never counted as your zero.
Can we show the proof?
For you to review
Shown in the report for you to look at. It costs no points and is not on your to-do list.
Counts, with a fixIt counts in the score, and the fix names the page, the rule or the number behind it.
You marked it intentional
Kept on a separate list and not reported as new again. The score stays as measured.
What counts as our failure, not yours
- A security service on your site, Cloudflare for example, blocked our tool. We do not then say your robots.txt or sitemap is missing.
- An AI model we use failed, or our tool could not open your pages and could only guess from their web addresses.
- We ran out of time, for example while looking for a connection an AI agent could use.
- We cannot search Google in your country. Our data provider has no Google results for Russia or Belarus, so we do not search the US instead.
- An assistant failed or sent back an empty answer. That is not the same as an answer that leaves you out.
From what we read to the number
Each part of the score is worked out from something we read: your site, the assistants' answers, Google and Bing, Google Maps and the connections an AI agent could use. Two of these need a word. robots.txt is a small file on your site that tells robots which pages they may read. A sitemap is a list of your pages that you publish for search engines. The same reads feed seven more checks that appear in your report and do not change the number.
- 16AI Share of VoiceRead from: Assistants' answers
- 13Retrieval ReadinessRead from: Your homepage, A sample of your pages, Your sitemap
- 11Search PresenceRead from: Google results, Bing's list of sites
- 12AI Identity & CrawlabilityRead from: Your homepage, robots.txt, Your sitemap
- 10Technical & PerformanceRead from: Google's free speed test
- 6AI SentimentRead from: Assistants' answers
- 6Mention PositionRead from: Assistants' answers
- 6Competitor GapRead from: Assistants' answers
- 10Local Presence (Google Maps)Read from: Google Maps, Your homepage
- 10Agent AccessRead from: Connections for AI agents
- In the report, not in the score
- Are the assistants right about youRead from: Assistants' answers, Your homepage, A sample of your pages
- Where the answers come fromRead from: Assistants' answers
- Named at each step of buyingRead from: Assistants' answers
- The narrower searches behind a questionRead from: Your sitemap, A sample of your pages, Google results
- YouTubeRead from: Your homepage, Google results
- Ready for AI shoppingRead from: A sample of your pages, robots.txt
- Dated pagesRead from: Your homepage, A sample of your pages, Your sitemap
Judged by the kind of business you are
An AI assistant expects different pages on different sites. A software company is expected to have pricing and comparison pages. A plumber is not, because customers call a plumber instead of comparing plans. So we first work out what kind of business the site belongs to, and then judge it only on the page types that kind needs. If you have such a page anyway, it still counts in your favour. If we cannot tell what kind of business you are, every page type applies.
Trade: a plumber, an electrician, a cleaner. Venue: a restaurant, a salon, a gym, a hotel. Firm: an accountant, a lawyer, an agency. Online shop: physical products sold for delivery. Industrial: machinery and materials sold to other businesses. Media: a publication, a blog, a directory or a course.
| Software | Online shop | Firm | Industrial | Trade | Venue | Media | Other or unclear | |
|---|---|---|---|---|---|---|---|---|
| Homepage | ||||||||
| Products | ||||||||
| Pricing | ||||||||
| Use cases | ||||||||
| Comparisons | ||||||||
| FAQ | ||||||||
| Works with | ||||||||
| About us | ||||||||
| Articles | ||||||||
| Definitions | ||||||||
| Case studies | ||||||||
| Types judged | 11 / 11 | 7 / 11 | 8 / 11 | 9 / 11 | 5 / 11 | 4 / 11 | 5 / 11 | 11 / 11 |
| Parts of the score by kind | ||||||||
| Google Maps | ||||||||
| Agent Access | ||||||||
Agent Access counts only when it helps you
AI agents are starting to do things for people: find a product and buy it, book a table, use a software service. Picture a customer asking ChatGPT to order a birthday cake from a bakery that sells on Shopify. The agent can only do it if the business offers a connection point for agents. Usually that is an MCP server, a standard way for an agent to search, order or book on a site by itself, or one the shop platform provides. This part checks whether you have one and how well it works.
Most businesses have none yet, and the standard is still settling. So this part counts only when it would raise your score: your result on it, as a share of its 10 points, has to be at least as good as your result on everything else. A shop with a good connection gains points. A shop with none loses nothing, and the report shows how many points a good one would add.
One exception always counts: a connection that secretly gives the AI instructions the customer cannot see. That is a real risk to anyone who connects an agent to it.
What “mentioned” means
We ask each question several times, because assistants answer differently each time. A question counts as naming you if any of those answers names you. Every number in the report uses this one rule, and the report also says how often the answers agreed.
Competitors are counted answer by answer. Ask an assistant for the best plumber in Austin: if it lists five firms and you are one of them, you have kept that answer. You lose an answer only when it names a competitor and leaves you out.
A competitor is a business of the same kind for the same customer. If you run a directory of tools, your competitors are other directories, not the tools you list.
Only what was measured counts
When we cannot check a part, we leave it out and divide by what is left. An earlier version of this scorer counted a missing check as zero, so when Google's speed test was down, sites lost twelve points for a fault that was not theirs.
So the total divides by the parts that actually ran. Say Google's speed test (10 points) was down and the assistants never named the business, so there was no tone (6 points) to judge. The score then divides by 84, not by 100, and the report says which two parts are missing and why.
That is also why the report shows how much of the score was measured. A 72 measured on everything and a 72 measured on three quarters of it are different results, and the report tells you which one you have.
One problem is charged once
If no assistant names a brand, three parts could all punish it at once: how often it is named, whether it is listed first, second and so on, and how often competitors take its place. Charging all three would push an ordinary result towards zero for a single fact.
So the first of them takes the full cost and the report names it as a critical problem. The other two report it and step out of the score.
An answer that names no business at all does not prove you have no competitors either, so it earns you no points.
Seven checks in the report, not in the score
The full report has seven more checks that do not change your number. They answer questions an owner asks: do the assistants get my prices right, which sites do they quote, do they know my products. Each one shows its proof, and its fixes are suggestions that carry no points. If one of these checks fails on our side, the report says so and draws no conclusion from it.
Are the assistants right about you
Assistants' answersYour homepageA sample of your pages
- What it checks
- What the assistants say about your business, compared with your own site: prices and plans, what you sell and what it does, where you are, and who founded or runs the business when your site says so. For example, an assistant quoting a monthly price your pricing page does not show.
- Where we look
- The answers that named you, plus one direct question to each assistant about your business, compared with your homepage and up to 4 of your product, pricing, about or FAQ pages.
- The proof you see
- The assistant, the question and its exact words, and next to them your page and its exact words. If a quote cannot be found word for word in the answer or on the page, we drop it.
- What it cannot do
- An assistant answering from memory changes only when its model is retrained. A fact buried deep in a long page may come out as not on your site. If your site does not mention something, we say so and do not call the assistant wrong.
Where the answers come from
Assistants' answers
- What it checks
- Which websites the assistants quote when they answer your customers' questions, how often, and which of them mention a competitor and never you. Those are the places worth getting mentioned.
- Where we look
- The answers we already collected from assistants that show their sources. No extra questions are asked.
- The proof you see
- Each site with the exact pages the assistants quoted and the questions they were answering.
- What it cannot do
- Full audit only. Assistants that do not show their sources are left out. We sort sites into types, such as review sites or forums, from a short list, so many land in other instead of being guessed.
Named at each step of buying
Assistants' answers
- What it checks
- How often you are named at each step a customer goes through: looking for the best option, trying to solve a problem, comparing two options, and checking reviews before buying. A brand can be named in every comparison and in no problem question.
- Where we look
- Each question is tagged with its step when it is written. If a question has no tag, a fixed word pattern in four languages assigns one.
- The proof you see
- Each step lists its questions. A step is shown only when at least two of its questions got an answer.
- What it cannot do
- Reports made before this check existed show nothing here. The tag is an AI's reading of the question, and the word pattern is rough.
The narrower searches behind a question
Your sitemapA sample of your pagesGoogle results
- What it checks
- When Google answers a question with AI, at the top of the results or in its AI Mode chat, it first breaks the question into several narrower searches. Google calls this query fan-out. For 3 of your customers' questions, an AI writes those narrower searches and we check whether a page of yours answers each one.
- Where we look
- The list of your pages, their titles and the links on your homepage, plus Google's top 10 results for up to 10 of those searches in your country.
- The proof you see
- The page of yours that answers the search, or your place in Google's results and the page that ranks. Two 2026 studies by companies that sell tools for ranking on Google point the same way: Ahrefs found that 38% of the pages Google's AI answers cite rank in the top 10 for the question itself, and Surfer found that pages ranking for the narrower searches were about 49% likelier to be cited.
- What it cannot do
- The narrower searches are an AI's guess, not Google's own list. We judge a page by its title and address, not by its full text. If your site has more pages than we list, we may have missed one, and the report says so.
YouTube
Your homepageGoogle results
- What it checks
- Whether your business has a YouTube channel, and which videos Google shows for your customers' questions.
- Where we look
- A YouTube link on your homepage, or else a Google search of YouTube for your brand, in your country.
- The proof you see
- Your channel's address and where we found it; up to 5 videos per question, marked when they name you. Ahrefs, a company that sells tools for ranking on Google, found that 18.2% of the pages Google's AI answers cite from outside the top 100 results are YouTube videos.
- What it cannot do
- We see YouTube through Google's results, not through YouTube's own search. A video about you on someone else's channel does not count as your channel.
Ready for AI shopping
A sample of your pagesrobots.txt
- What it checks
- For an online shop: whether your product pages carry what a shopping assistant quotes, which is the name, the price with its currency, whether it is in stock, a product code such as a barcode number, the brand and a picture. Also whether a product feed, the file a shop sends to Google or ChatGPT with all its products, can be found.
- Where we look
- The product details in the code of your sampled pages, and a few requests to the usual places a shop publishes its feed.
- The proof you see
- Each product page's address with the values exactly as written, and the feed's address when we find one.
- What it cannot do
- We usually read one or two product pages. A feed you send privately to Google Merchant Center, Google's service where shops upload their product lists, or to OpenAI is invisible to us, so a feed we could not find is something to review, never a fix.
Dated pages
Your homepageA sample of your pagesYour sitemap
- What it checks
- Whether each page we read says when it was last updated, and whether that date is 12 months old or more. Assistants quoting a price or a comparison prefer a page that says when it was true.
- Where we look
- A date printed next to words such as updated or published, in four languages; the date in the page's code; the hidden date your website sends along with each page; and the date in your sitemap.
- The proof you see
- The newest date we found, where we found it and its exact words.
- What it cannot do
- A date with no such word next to it is ignored, and so is a date in the future or one that could be read two ways, such as 03/05. A server that always sends today's date makes every page look fresh.
You can mark a problem as intentional
Some of what we flag is a decision you made. A shop may close its customers' account pages to crawlers on purpose, and it should not hear about it every week.
First you prove the site is yours, for example by confirming an email address on its domain. Then you can mark the problem as intentional and write why. It moves to its own list in your report, with your name and the date, and stops coming back as new in our weekly letters and on your page in the Atlas, our public catalogue of AI tools. If a new rule of the same kind appears later, it is a new problem and we report it.
Your mark does not change the score. It only records your decision.
The detail
The rest of the page is for anyone who wants to check the arithmetic: how many questions the score rests on, how sure it is, and how each part is worked out inside.
How many answers the score rests on
The biggest part of the score comes from asking assistants real buying questions and reading what they say. A percentage means little without knowing how many answers it comes from, so every report prints its own sample.
- Questions
- Up to 20 per audit, written from your own site: what you sell, who you serve, which problems you solve. A question that contains your brand name is dropped, because the answer would name you anyway.
- Answers per question
- Seven on a full audit for the first 8 questions, the ones about buying your product, each asked of the assistant's top model. The rest go to the same company's cheaper model, 3 times each, the same as a person with a free account gets. Assistants answer differently every time, so a single answer cannot tell a real change from chance.
- Assistants
- Claude, ChatGPT, Gemini and Perplexity, each through its own company's model. The report names the model behind every answer and says how many assistants actually answered.
- Memory and search
- We ask each question twice: once with the assistant answering from what it learned, and once with its web search switched on. They are reported separately because they are different facts.
- The combined figure
- When both kinds of answer came back, how often you are named is 60% from the answers with search and 40% from the answers from memory. Search answers are what people see today, and memory still matters for chats that never search. If one kind failed completely, we use only the other, because our failure is not your 0%.
- Failed answers
- An answer the assistant refused, ran out of time on or sent back empty is left out and counted separately. It is not scored as an answer that left you out.
Assistants in the audit
- Claudesearches
- ChatGPTsearches
- Geminimemory only
- Perplexitysearches
Gemini answers from memory only: through the connection we use, it does not return the pages it relied on.
The free scan uses the same parts with one assistant, one answer per question and no search. Its number is a rough guide; the full audit measures.
How sure the number is
Every figure here is a share worked out from a few dozen answers on one afternoon. Zero out of six and zero out of six hundred both print as 0% and mean very different things, so the report shows a range around how often you are named: the values the true figure most likely falls between, with 95% confidence.
We work out that range with a method that stays accurate on small samples, the Wilson interval. The textbook method claims no uncertainty at all when the result is exactly 0% or 100%, which is when the sample usually tells you the least.
Asking one question seven times is not seven separate facts about the world. Seven answers to each of twenty questions are twenty observations, not a hundred and forty, and the range is worked out over questions for that reason. Counting every answer separately would make the range about two and a half times too narrow.
The confidence label takes the weaker of two things: how much of the score was measured, and how wide that range came out. A sample of six questions keeps the label low even when every part was measured.
What the confidence label means
- High
- At least 90 of the 100 points were measured, and the range around how often you are named is no wider than 15 points.
- Medium
- At least 70 points measured, and the range no wider than 30 points.
- Low
- Anything below that. The number shows a direction, and the report says so.
- Google: AI features and your website (the guidance this audit does not try to outsmart)
- SE Ranking: 300,000 domains, no link between having llms.txt and how often AI cites a site
- seoClarity, October 2025: 362,000 searches, 56% of the pages Google's AI answers cite come from the top 20 ordinary results
- Schulte et al. 2026, and the 2026 critical survey of generative-engine optimization, on margins of error and on repeated answers
From points to a verdict
The word next to your number comes from where the number falls, not from any single problem.
- 70 and above
- Looking good
- 45 to 69
- Needs attention
- Below 45
- Critical issues found
Two or more critical problems move the verdict down one step. One does not, so a single problem cannot turn a site scoring 90 into a critical case, which is what an earlier rule did. If your homepage is built entirely in the browser and arrives empty, the three problems that causes (no business labels, no description, no preview) stay on the list but count as one cause and do not move the verdict.
Inside Retrieval Readiness
This part asks whether your site has the pages an assistant would want to quote, and whether your homepage says clearly enough what you are for an assistant to put you in the right category.
Having the right page types is worth 60 of this part's 100 points. A clear homepage, itself scored out of 100, gives the other 40. A page type counts as missing only if we went through your whole sitemap, or every link on your homepage, and found no page of that type. If your sitemap is longer than what we read and no such page turned up, the type is left out of the 60 points instead of counting as missing. A page we found only by its address still counts as present.
The eleven page types
- A homepage that says who you are, what field you work in and who you serve
- A page for each product or service, saying what it does and who it is for
- Prices, plans or how you charge
- Pages about specific problems you solve or industries you serve
- Comparisons with other options, and alternative-to pages
- Questions and answers
- What your product works with
- Team, mission, certificates, testimonials
- Articles, guides and your own research
- Clear definitions of the terms in your field
- Case studies, test results, your own data
Five things a clear homepage says
Each is judged clear, vague or missing. Clear earns 20 points, vague earns 8, missing earns nothing, so five clear answers make 100 and five vague ones make 40.
- What field you are in, said outright
- Who your product or service is for
- How you differ from the alternatives
- Visible proof: numbers, case studies, references
- What you actually do, and how
Inside AI Identity
Read from your site's root and scored out of 100, which then becomes this part's share of the total. A file we could not read is left out; it does not count as missing.
The crawlers that count, and what each is worth
These are the crawlers that fetch pages when an assistant searches the web, so they decide whether your site can appear in an answer. Together they are worth the 25 points of crawler access.
Allowed earns the full points. Allowed except for some sections earns 70% of them, and only when the closed section holds a page you publish in your sitemap or link to. Other closed sections are listed for you to review and cost nothing. Blocked earns nothing.
The crawlers that collect pages to train AI models, such as GPTBot, ClaudeBot, Meta-ExternalAgent and Bytespider, are reported and carry no points at all. We suggest letting them in: when an assistant does not search, it answers from what its model learned, and blocking Google-Extended, Google's switch for AI training, takes your site out of the Gemini app (Google's AI answers in search are not affected). The choice is yours, so the score does not take points for it.
Inside Technical & Performance
One run of Google's PageSpeed test on your site, weighted: speed 40%, basic search setup 25%, accessibility 20%, good practice 15%.
A category the test did not return is left out rather than scored zero, and the rest are divided by their own share. A test that returned nothing at all is treated as no measurement, not as a site that failed every check. A slow result on mobile becomes a fix with its measured score and page; it is never a critical problem on its own.
Inside Search Presence
The same buying questions, leaving out any that contain your name, searched on Google in your own country and checked against Bing's list of sites.
Google is worth 8 points: 70% of them for how many questions show your site in the top 10, 30% for how many show it in the top 3. Bing is worth 3, for having your site at all.
We count only the searches that actually came back, not every search we sent. The report says you are missing from Google's top 10 only when at least 3 searches came back, and never about a result nobody read.
What the score does not measure
Here is what the score leaves out.
- It is not a ranking. No index of every company exists to give you a place in; the score describes your site against the questions your own content suggests.
- It is a snapshot. Asking each question several times separates a real change from an assistant's own variation, but it does not turn one afternoon into a trend. Running the audit again does that.
- It does not see personal answers. Assistants change their answers by account, history and location, and an audit sees one anonymous view.
- The questions come from your site. A site that describes itself vaguely gets vaguer questions, and its score shows that instead of making up for it.
- It does not score files such as llms.txt or ai.json, new formats meant to summarise a site for AI. They are reported and carry no points, because there is no evidence yet that they make assistants quote a site more.
- It does not score whether you let AI companies train on your pages. Opting out has no measured effect on being quoted.
FAQ
- Why is my free score different from the paid one?
- The free scan asks one assistant, asks each question once and skips the answers with web search, so it rests on far fewer answers and its range is much wider. The parts and their points are the same. The difference is the size of the sample, and the report says so.
- Why did my score change when nothing on my site changed?
- Assistants do not give the same answer twice. That is why a full audit asks its main questions seven times and shows a range: a change inside the range is the assistant's own variation, and a change outside it is worth acting on.
- Will having no MCP server lower my score?
- No. Agent Access counts only when it raises your score. The report shows how many points a good connection would add. The only one that can lower it is a connection that secretly gives the AI instructions the customer cannot see.
- You flagged something I did on purpose. What can I do?
- Prove the site is yours, then mark the problem as intentional and write why. It moves to a separate list and stops coming back as new. Your score stays as measured.
- Do you sell the thing you score?
- Yes, which is why we publish the arithmetic. Every point value, threshold and formula on this page is the one the code uses. If a part looks wrong to you, you can check it against the numbers here.
- Why does llms.txt not earn points?
- SE Ranking looked at 300,000 domains and found no link between having the file and being quoted by AI. It may still be a useful list of your pages, but it is not evidence of visibility, so scoring it would invent a signal.
- Can I see the answers themselves?
- Yes. The report shows a typical answer to each question, which assistant gave it, and whether you and which competitors were named.
Check your own site
The free scan uses the same parts with a smaller sample. You see the score without an account.
Start the free scanCookies
Before we count this visit
Analytics and ad measurement tell us which pages and which of our ads do their job: Google Analytics and Google Ads, the Meta pixel, Yandex Metrica, and one first-party id of our own. Nothing is stored until you answer, and declining changes nothing about the site.