AI Visibility Score: How It's Calculated and What It Can't Tell You

An AI visibility score is a 0 to 100 number that estimates how often, and how prominently, AI assistants such as ChatGPT, Perplexity, Gemini, Claude and Google’s AI Overviews mention or cite your brand when people ask questions in your category. It is calculated by running a fixed set of prompts, scoring each answer and dividing the points earned by the maximum possible. It measures presence in answers, not traffic or revenue.

That definition hides most of what matters. Every tool picks its own prompts, platforms and weights, so scores from two vendors are not comparable. The number also moves on its own because AI answers vary from run to run. This guide shows exactly how the score is built, works one through with real arithmetic, and explains how to tell whether a change in the score is real and whether it is worth anything.

Two different things are called an AI visibility score

Search for the term and you will find two kinds of product using the same name. They answer different questions.

A readiness score is a checklist result. A page can pass every check and still never be recommended, because recommendations depend on what the rest of the web says about you and on which competitors exist. Use readiness scans to remove blockers. Use answer-based scores to see where you stand. The rest of this article is about the answer-based kind, since that is what most people mean.

What an AI visibility score measures

Most answer-based scores combine some or all of these signals:

Some vendors add derived measures such as consistency (how reliably you appear when the same question is phrased differently or asked again) and authority signals. These are all variations on the same input: a set of answers, each marked for the signals above.

How an AI visibility score is calculated

1. Pick the prompts. Write 20 to 30 unbranded questions a buyer would ask before choosing a product like yours: “best [category] for [use case]”, “[competitor] alternatives”, “[brand A] vs [brand B]”, and problem questions such as “how do I fix [pain point]”. Branded prompts (“what is [your brand]”) inflate the score and tell you little. Assistants often split one question into several searches behind the scenes (query fan-out), so long-tail phrasings from your keyword research make good prompts.
2. Pick the platforms. Choose the assistants your buyers use. Covering more platforms gives a broader view but makes the score harder to move.
3. Run every prompt in a fresh session. A new chat each time, so earlier answers do not shape later ones. Keep location and language fixed.
4. Score each answer. Award points per signal: mentioned, cited, near the top, positive. The weights are a choice, and that choice changes the result, as the example below shows.
5. Normalize. Add the points earned and divide by the maximum possible points.

Two simpler ratios sit alongside the score and are worth tracking on their own:

• Presence rate = answers that mention you ÷ total answers.
• Share of voice = your mentions ÷ all brand mentions across the same answers.

Worked example: one set of answers, three different scores

The numbers below are illustrative, for a made-up project management tool. They show the arithmetic from start to finish.

Weighting A: every signal worth 1 point

Each answer can earn 3 points (mention, citation, top three). Maximum = 80 × 3 = 240. Earned = 30 + 12 + 18 = 60. Score = 60 ÷ 240 × 100 = 25.

Weighting B: citations worth 3 points

A tool that values links more might weight a citation at 3. Each answer can now earn 5 points. Maximum = 80 × 5 = 400. Earned = 30 + (12 × 3) + 18 = 84. Score = 84 ÷ 400 × 100 = 21.

Weighting C: presence only

A tool that only checks whether you are named reports the presence rate: 30 ÷ 80 × 100 = 37.5.

Share of voice is 30 ÷ 200 = 15%, whichever weighting you use.

How to measure it yourself, without a tool

A manual baseline takes an afternoon and teaches you more about how assistants treat your category than any dashboard.

1. Write your prompt list and freeze it. Save it with a date.
2. List three to five competitors you expect to see.
3. Run each prompt in a fresh chat on each platform. For every answer, log: mentioned (yes/no), cited with a link (yes/no), position in any list, sentiment, every competitor named, and every source domain cited.
4. Calculate presence rate, share of voice and your weighted score with the formula above. Write down the weights you used.
5. Repeat on the same day of the week or month. The first run is your baseline: measure before you change pages, otherwise you will not know what moved the number.

The source-domain column is the most useful output. It shows which review sites, lists, forums and publications the assistants lean on in your category, which is where your improvement work should go.

Why your score jumps around, and when a change is real

AI answers are not fixed. Ask the same question twice and you can get a different list. Models are updated, the web pages they retrieve change, and vendors re-run prompts on their own schedules. Small prompt sets turn that noise into large swings.

The arithmetic is simple. With 20 prompts on one platform, each prompt is worth 5 points of presence rate. If two answers flip from “not mentioned” to “mentioned”, your score rises 10 points without anything real happening. With 80 answers (20 prompts on 4 platforms), each answer is worth 1.25 points. Run every prompt three times and record the share of runs that mention you (for example, 2 of 3 = 0.67), and each run is worth about 0.4 points.

Debugging a sudden drop

Before you rewrite pages, rule out the causes that have nothing to do with your content:

What the official docs say vs what the top results claim

Several guides and audit tools treat an llms.txt file or extra schema as a quick win, and some readiness scores deduct points without one. Google’s own guidance disagrees for its AI features. Its documentation on AI features says there are no additional requirements or special optimizations to appear in AI Overviews or AI Mode, that you do not need new machine-readable files, AI text files or markup, and that no special schema.org structured data is required. A page needs to be indexed and eligible to show in Search with a snippet. The same page notes that traffic from these features is included in your overall Search Console traffic, so you cannot split it out there.

Crawler access does matter for ChatGPT. OpenAI’s crawler documentation says OAI-SearchBot is what surfaces websites in ChatGPT search, and that sites opted out of it will not be shown in ChatGPT search answers. It is a separate bot from GPTBot, which collects training data, so you can block one and allow the other.

The practical reading: structured data and llms.txt will not hurt, but do not expect them to move an answer-based score on their own. Indexability and crawler access are the real technical prerequisites.

How to improve your AI visibility score

• Earn mentions where assistants already look. Your source-domain log shows the comparison articles, review sites, Reddit threads and publications cited for your prompts. Getting included there is the most direct lever for unbranded prompts. On forums, disclose your affiliation and answer the question; promotional posts get removed.
• Answer the question at the top of the page. Lead key pages with a two or three sentence answer, then use clear headings, lists, tables and short FAQ blocks. This makes passages easy to extract and quote.
• Publish something only you have. Original data, pricing transparency, real comparisons and first-hand experience give an assistant a reason to cite you instead of a generic summary.
• Say clearly who you are, what you do and who you serve. On the homepage and product pages, in plain words. Ambiguity about your category makes it harder to be named in it.
• Keep comparison and alternatives pages current. “[Competitor] alternatives” and “X vs Y” prompts are where shortlists form.
• Fix the technical basics. Indexable pages, allowed search crawlers, fast server responses, and content that renders without JavaScript tricks.

For the wider strategy behind these tactics, see what AEO and GEO mean in practice.

AI visibility score vs SEO metrics

The two are related because many assistants retrieve indexed web pages before answering, so strong organic pages often feed AI answers. But a page can rank first and be absent from answers, and a brand can be named in answers with no page ranking at all. Our comparison of AI search vs SEO covers the measurement differences in more detail.

Tools that report a score

Answer-based scores come from prompt-tracking platforms, including AI modules inside SEO suites (Semrush, Ahrefs Brand Radar, SE Ranking) and dedicated trackers (such as Profound, Peec AI and Otterly). Readiness scores come from site scanners and browser extensions. Before you pay for one, check:

• Which platforms it queries, and whether that matches where your buyers ask.
• Whether you can see the exact prompts, the raw answers and the weights behind the score.
• How often it re-runs prompts and whether it averages several runs.
• Whether it shows competitor share of voice and cited source domains, not just your number.
• Whether it records model and method changes so you can re-baseline.

For a side-by-side on how visibility trackers differ from conversion tracking, see SEOConversion vs AI visibility trackers.

What a good score does not tell you

A score is a diagnostic, not an outcome. It cannot tell you whether anyone clicked, whether those visitors converted, or which page they landed on. Many answers name a brand without a link. Some assistants send visits with no referrer, so those sessions appear as Direct in your analytics. A score can rise for months while revenue from AI stays flat, and the opposite can happen too.

From score to revenue: an illustrative example

Suppose the brand from the earlier example raises its score (weighting A) from 25 to 34 over a quarter. Was it worth the effort? The score cannot say. AI referral visits and conversions by landing page can. The numbers below are illustrative.

Two things stand out. The blog guide gets the most AI visits but produces the least value, so a visibility gain there is worth less than it looks. The comparison page converts at 5% (6 ÷ 120) and drives two thirds of the value, so the prompts that cite it are the ones to protect. If you have not set a value per conversion yet, use this guide to calculating conversion value.

This is the half of the picture SEOConversion covers. It does not produce a visibility score or track prompts. It records conversions and their value from identifiable ChatGPT, Perplexity, Claude, Gemini and Copilot referrals, by landing page, and leaves visits with no referrer as Direct. The full setup is in our guide to AI conversion tracking.

Frequently asked questions