A generative engine optimization calculator is a worksheet that turns saved AI answers into four ratios: prompt coverage, mention rate, citation rate and share of voice. No formula can size your opportunity without a fixed prompt set, a count of engines and runs, and the answers themselves. This guide gives the formulas, a worked example and the limits.
Key takeaways
- Build the calculator around a fixed prompt set. Change the set and every ratio changes, so freeze it for each reporting period.
- Track four ratios: prompt coverage, mention rate, citation rate and share of voice. Each has a different denominator, and mixing them up is the commonest error.
- Report the number of answers behind each ratio. A percentage from 30 answers is a hint, not a result.
- Size the opportunity as the list of prompts where competitors are named and you are not, not as a made-up market figure.
- Do not use a one-number "GEO score" from a URL scanner as a performance measure. It grades your page, not what engines say.
How do I calculate my generative engine optimization performance?
You calculate it from observed answers. Choose the prompts your buyers ask, run each one on every engine you care about, repeat each run a few times because answers vary, save the text and the cited sources, then count. Everything below is arithmetic on that table.
Four inputs define the sample:
- P: prompts tracked.
- E: engines queried.
- R: runs per prompt per engine.
- N = P × E × R: total answers collected.
Every ratio in the next section is a count taken from those N answers. If you want help choosing P, read how to choose prompts to track and prompt research for AI search. The free AI Prompt Finder lists the questions buyers ask in your category, tagged by intent.
What metrics go into a GEO calculator?
Four ratios cover most decisions. Their formulas are plain, and published AI share of voice guides use the same logic: RankScope, October 2026, defines share of voice as your brand citations divided by total citations across tracked brands, multiplied by 100. The table sets out all four with the denominator that matters.
| Metric | Formula | Denominator | What it hides |
|---|---|---|---|
| Prompt coverage | Prompts where your brand is named at least once ÷ P | Prompts | How often you are named within a prompt |
| Mention rate | Answers naming your brand ÷ N | All answers | Whether competitors are named more |
| Citation rate | Answers citing your domain ÷ answers that show sources | Answers with sources | Engines that show no sources are excluded |
| Share of voice | Your mentions ÷ mentions of all tracked brands | Brand mentions | Depends on which competitors you list |
A fifth and sixth are optional: average position in a list answer, and sentiment. Add them only once the four above are stable.
Why mention rate and citation rate are different numbers
A brand can be named without its site being linked, and a page can be cited while the answer names someone else. Citation versus brand mention explains the split. Your calculator should keep two columns, because the fixes differ: mentions come from what the wider web says about you, citations from which of your pages engines can read and trust.
Choose the denominator on purpose
Alex Birkett's guide to AI share of voice gives the basic version as responses mentioning your brand divided by total responses for the prompt set (Alex Birkett, 2026). That is what this worksheet calls mention rate. The RankScope version above divides by all brand mentions, which is what this worksheet calls share of voice. Both are valid. They are not the same number, so label which one you report.
How do I run the worksheet step by step?
- Freeze the prompt set. Write 30 to 60 prompts across stages: category discovery, comparison, and problem-led. Record the date and do not edit them mid-period.
- Name the competitor set. List the brands whose mentions count in the share of voice denominator. Add alternate spellings of every brand.
- Pick engines and runs. Use the engines your buyers use. Run each prompt at least three times, because the same prompt can return different answers.
- Save the answers. Keep the full text, the list of brands named, and the cited URLs.
- Fill the table. One row per answer, with columns for engine, prompt, run, brands named, domains cited.
- Compute the four ratios for the whole set and per engine.
- List the gaps. Filter for answers where a competitor is named and you are not.
- Repeat on the same schedule and compare only like with like.
What does a worked example look like?
The numbers below are invented to show the arithmetic. They are not a benchmark, and no real brand is described.
Suppose a team tracks P = 40 prompts on E = 4 engines with R = 3 runs. That gives N = 40 × 4 × 3 = 480 answers.
| Input (hypothetical) | Count | Calculation | Result |
|---|---|---|---|
| Prompts naming the brand at least once | 22 of 40 | 22 ÷ 40 | Prompt coverage 55% |
| Answers naming the brand | 120 of 480 | 120 ÷ 480 | Mention rate 25% |
| Answers showing sources | 300 | Denominator for citation rate | n/a |
| Answers citing the brand's domain | 54 of 300 | 54 ÷ 300 | Citation rate 18% |
| Mentions: brand 120, rival A 180, rival B 90, rival C 60 | 450 total | 120 ÷ 450 | Share of voice 26.7% |
Read it as a team would. Coverage of 55% means 18 prompts return no mention at all, and those 18 are the opportunity list. A mention rate of 25% alongside a citation rate of 18% suggests the brand is named more often than its pages are used, which points to content and entity work rather than outreach alone.
Put a margin on the percentage
With N = 480 and a 25% mention rate, the standard binomial margin at 95% confidence is about 1.96 × √(0.25 × 0.75 ÷ 480), roughly ±3.9 percentage points. That treats each answer as independent. Runs of the same prompt are not independent, so the real margin is wider. A move from 25% to 27% is noise. A move from 25% to 35% across two periods with the same prompt set is worth investigating.
How do I size the opportunity without inventing market numbers?
Public AI prompt volume data is thin, and a calculator that multiplies by a guessed volume gives false precision. Size the opportunity from what you observed instead, in three layers:
- Gap prompts: prompts where at least one competitor is named and you are not. Count them and rank by buyer stage.
- Gap engines: engines where your mention rate is below your overall rate. Engines often disagree, as why AI engines disagree sets out.
- Gap sources: domains cited in answers that name competitors but never name you. These are the pages and publications to earn a place on.
If you want to convert this to revenue, use your own numbers: the share of your pipeline that says it used an AI assistant, taken from your CRM or a signup question. That is yours to measure; do not borrow someone else's. See measuring AI influence beyond referrals and how to report AEO to a CFO.
Which inputs predict the numbers you are calculating?
The calculator measures outcomes. To know what to change, look at published research on drivers. The GEO paper by Aggarwal and co-authors, presented at KDD 2024, reported that its methods could boost visibility by up to 40% in generative engine responses (ACM, 2024). That is a ceiling from a specific benchmark, not a promise for your site.
For Google AI Overviews, Ahrefs' study of 75,000 brands found branded web mentions had the strongest link to AI visibility, according to Ahrefs' tracker page, checked November 2026. LSEO's summary of the study reports a correlation of 0.664 for branded web mentions against 0.218 for backlink count (LSEO, October 2026). Correlation is not cause, but it tells you which column of your gap-sources list to work first. More on this in which mentions drive AI visibility.
Google's own guidance frames the work plainly:
"optimizing for generative AI search is optimizing for the search experience, and thus still SEO." Google Search Central, May 2026
The practical reading: keep your technical SEO and content quality sound, and use the calculator to see whether it shows up in answers.
Is a free GEO score the same as a calculator?
No. Some free pages call themselves a GEO calculator. Vibe Eval's page, as checked on 16 November 2026, takes a URL and returns a score across 8 dimensions of how citable the page is (Vibe Eval, November 2026). That grades a page's readiness. It does not ask any engine a question, so it cannot give prompt coverage, citation rate or share of voice.
| Question | Page-readiness score | Answer-based worksheet |
|---|---|---|
| Input | One URL | Prompts, engines, runs, saved answers |
| Tells you | Whether a page is easy to read and quote | Whether engines name and cite you |
| Competitor view | No | Yes, through share of voice |
| Changes over time | Only when you edit the page | Whenever answers change |
| Use it for | Fixing a page | Judging performance |
Use both, in order. The free Noma AI SEO Checker runs 30+ checks on whether ChatGPT, Claude, Perplexity and Google's AI Overviews can read and quote a page, so you can fix readiness before measuring answers. For crawl access specifically, the AI Crawler Checker shows whether GPTBot, ClaudeBot and others are allowed in robots.txt.
Where do the inputs come from at scale?
A spreadsheet works for 40 prompts and one engine. At 40 prompts, 4 engines and 3 runs a month you are saving 480 answers by hand, and the sample shrinks the first time someone skips a week. The practical limit is collection, not arithmetic.
Noma by Nurdd asks 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, the questions your buyers ask, every day. It shows where your brand is named, where competitors are named instead, and which pages the answers draw on. Those are the raw counts this worksheet needs for coverage, mention rate, citation rate and share of voice. Its help article on how visibility is calculated explains its own definitions.
See it on your own brand: book a demo or start the 3-day free trial. A card is taken at signup so billing can start on day 3; cancel before then and nothing is charged. For a no-signup first look, the free brand audit asks ChatGPT, Perplexity and Gemini 10 buyer questions.
What mistakes make a GEO calculation misleading?
- Changing the prompt set between periods. You are then comparing two different questions.
- One run per prompt. Answers vary, so a single run can show you present or absent by chance. See AI citation stability.
- Branded prompts only. Prompts that include your name inflate mention rate. Keep them in a separate bucket.
- Averaging engines without showing the spread. A 25% overall rate can be 60% on one engine and 5% on another.
- A thin competitor list. Leaving out a rival inflates your share of voice.
- Ignoring wrong answers. Being named with a wrong fact is not a win. Check with an AI brand fact check.
- Reporting percentages with no N. Always print the answer count beside the ratio.
How often should I recalculate?
Collect daily or weekly if you can, report monthly. Compare the same prompt set across the same engines, and mark any date when a model release or site change could have shifted answers; new AI model releases and brand visibility covers what to check. For a broader measurement plan, see how to measure AI share of voice across engines.
How we researched this
We reviewed Google Search Central documentation, the peer-reviewed GEO paper, Ahrefs' own page on its 75,000-brand study and a secondary summary of it, and three published articles on AI share of voice and GEO scoring. We searched the primary keyword and read the top-ranking results for what they cover and omit. The worked example is hypothetical arithmetic, not data. Facts about third-party pages were checked on 2026-11-16.
What to do next
Copy the table into a spreadsheet, freeze 30 to 60 prompts and collect your first month of answers. Then compare the gap list with your content plan. To have the answers collected for you across 13 AI engines, book a demo or start the free trial.
See what AI engines say about your brand
Free · no account · or start a 3-day trial
Questions people ask
How do I calculate my generative engine optimization performance?
Freeze a set of buyer prompts, run each on your chosen AI engines several times, and save the answers and cited sources. Then compute prompt coverage, mention rate, citation rate and share of voice from those saved answers. Repeat monthly with the same prompts so the comparison is fair.
What metrics go into a GEO calculator?
Four: prompt coverage (prompts naming you at least once divided by prompts tracked), mention rate (answers naming you divided by all answers), citation rate (answers citing your domain divided by answers that show sources) and share of voice (your mentions divided by all tracked brands' mentions). Always report the number of answers behind each.
Is there a free generative engine optimization calculator?
Free page scorers exist, but they grade a URL and do not query AI engines, so they cannot give share of voice or citation rate. You can build the answer-based version in a spreadsheet using the formulas in this guide. Noma's free AI SEO Checker tests whether a page can be read and quoted.
How many prompts and runs do I need for reliable numbers?
There is no universal threshold. A practical start is 30 to 60 prompts with at least three runs each per engine. Treat the binomial margin as a minimum, because runs of the same prompt are not independent, and avoid reading small moves as real change.
What is the difference between mention rate and share of voice?
Mention rate divides answers naming your brand by all answers collected. Share of voice divides your mentions by the mentions of every tracked brand. Mention rate shows how often you appear; share of voice shows how you compare with rivals.
Can a GEO calculator predict revenue?
Not reliably. Public AI prompt volume is thin, so any revenue forecast built on it rests on guesses. Use your own data, such as the share of leads who report using an AI assistant, and treat the calculator as a measure of visibility.
