Methodology

How flowgen measures AI visibility

Every number on this site comes from observable AI responses, your website and public sources. This page explains what is measured, how each score is calculated, and what the numbers can and cannot tell you.

Observed responses, dated runs, labelled estimates

What flowgen observes

Three rules apply to every measurement, on every plan.

Observable responses, not model internals

flowgen records what an AI platform actually answered to a question, as a user would see it. It does not claim to know how a model weighs sources or what it “thinks”.

Your prompt set, not a generic panel

Scores are calculated on the questions tracked for your brand and category. Two brands are comparable only when they are measured on the same questions, which is how competitor comparison works.

Dated runs, so change is measurable

Each collection is dated. Change is the difference between two dated runs on the same questions, never a single reading.

The measurement pipeline

Seventeen steps, in five stages. Each product on the platform reads from one or more of them.

Stage 1Questions

Prompt selection. Questions come from the buyer journey for your category: your own search and support data, the questions flowgen AI Demand finds, and the ones your team adds. Brand-free questions are included on purpose.
Topic clustering. Questions are grouped into topics so a score can be read per topic, not only per brand.
Intent classification. Each question is tagged by buying stage and intent: exploration, comparison, recommendation, specification, use case, after purchase.

Stage 2Answers

AI platform coverage. Questions are run on the platforms included in your workspace, such as ChatGPT, Gemini, Perplexity, Claude, Google AI Overviews, Google AI Mode, Microsoft Copilot and DeepSeek. Each report lists the platforms it covers.
Sampling. Answers vary between runs. Each question is asked more than once per run where the platform allows it, and the share is calculated over the sampled answers. Sample sizes are shown in the report.
Brand mention detection. Brand, product and competitor names are matched in each answer, including common variants, and the position of each mention is recorded.
Citation analysis. Every cited URL is recorded and classified by domain, source type (owned, earned, community, marketplace, video) and the page cited.
Competitor comparison. The same answers are read for the competitors you track, so rank and share use identical questions and runs.

Stage 3Meaning

Brand Knowledge analysis. Recurring phrases attached to the brand are grouped into themes and classified as aligned, missing, mixed, outdated, inaccurate or risk. This is what Brand Knowledge reads.
Claim verification. Factual statements in answers are extracted and checked against the verified facts in Brand Knowledge, then marked correct, outdated, incorrect or unverified. This is what Brand Knowledge reads.

Stage 4Website

Crawlability. Whether the crawlers used by AI platforms can reach your key pages, and what robots rules say.
Entity consistency. Whether the company, product names, addresses and key facts are the same across your site and the public sources AI cites.
Content coverage. For each tracked question, whether a page on your site answers it, partially answers it, or does not exist.
Source authority. How often your domain is cited compared with the other sources cited for your questions, and which external sources carry the most weight in your category.
Technical GEO readiness. Structured data, text versus image content, page freshness and the presence of retired pages still being indexed.

Stage 5Scores

Score normalization. Each score is scaled to 0 to 100 so that topics, platforms and competitors can be compared on one axis. The underlying counts stay visible in the report.
Change measurement. Before-and-after comparisons use the same questions, platforms and sampling on both dates. A change that follows an action is reported as an association, not as proof of cause.

The six scores

What each one measures, why it matters, how it is calculated and how to move it.

AI Visibility

Discover · flowgen AI Visibility

What it measures: The share of sampled answers to your tracked questions that name your brand.
Why it matters: If you are not in the answer, you are not on the shortlist. This is the first number buyers’ decisions depend on.
How it is calculated: Answers naming the brand ÷ sampled answers, per platform and per topic, then normalized to 0–100.
How to improve it: Answer the lost questions with a page that states the facts directly, and get those facts repeated by the sources each platform cites.

Citation Readiness

Discover · flowgen AI Visibility

What it measures: How often your own pages are cited as a source, and how citable your key pages are.
Why it matters: A brand can be named while a competitor’s page is cited. Citations are where the authority sits.
How it is calculated: Share of citations pointing to your domain across sampled answers, weighted with a page-level check for plain-text facts, structure and freshness.
How to improve it: Put the quotable facts in sentences and tables, date the pages, and retire old versions.

Content Coverage

Discover · flowgen AI Demand

What it measures: The share of tracked questions that a page on your site answers.
Why it matters: A missing page is the most common reason a competitor is named instead of you.
How it is calculated: Each question is matched to the best page on your site and marked covered, partial or missing. Covered ÷ tracked questions, normalized.
How to improve it: Create the missing pages in order of opportunity score; improve partial pages before creating new ones.

Source Authority

Understand · flowgen AI Visibility, Brand Knowledge

What it measures: How much weight the sources that mention you carry in your category, and how many independent sources confirm your key facts.
Why it matters: Engines repeat facts that several trusted sources agree on. One page saying it is rarely enough.
How it is calculated: Citation frequency of each source across your category’s questions, combined with how many of those sources mention your brand consistently.
How to improve it: Earn mentions in the sources the engines already cite, and correct the ones that describe you wrongly.

Competitor Gap

Discover · flowgen AI Visibility

What it measures: The questions where a tracked competitor is named and you are not, and how far their visibility is ahead of yours.
Why it matters: This is the shortest list of things to fix: each item is a question a buyer is asking today.
How it is calculated: Per question, competitor named and brand not named, counted and weighted by the question’s opportunity score.
How to improve it: Work the gap list from the top. The reason for each gap (missing page, weak source, wrong fact) is shown next to it.

GEO Readiness

Discover · flowgen AI Visibility

What it measures: Whether your site can be found, read and trusted: crawl access, structured data, entity consistency, text facts, freshness.
Why it matters: None of the other scores can move if the engine cannot read the page or finds two versions of the truth.
How it is calculated: A checklist of technical and consistency checks on your key pages, each pass or fail, weighted and normalized.
How to improve it: Fix the failed checks in the order the report lists them. Most are small changes to existing pages.

How numbers are labelled

Every figure on the platform and on this site carries one of these labels. Demo data is never mixed with observed data.

Label	What it means	Where you will see it
Observed	Counted directly from sampled AI responses or from your website on a dated run.	Mentions, citations, claims, crawl checks
Estimated	Derived from observable signals with a stated method. flowgen does not report exact prompt volumes or user counts for any AI platform.	AI demand, AI referral lift
Modeled	Produced by a scoring model that combines several observed inputs, such as the opportunity score.	Opportunity score, GEO Readiness
Relative index	A 0–100 scale for comparison between topics, platforms or brands. Not a percentage of anything outside your prompt set.	Score cards, trend lines
Sample data	Illustrative numbers for a fictional brand, used on this website to show how a product works.	Every demo window on this site

What flowgen does not claim

No exact prompt volume. How many people ask a question on an AI platform is not observable from outside; flowgen reports estimated demand as a relative index.
No guarantee of a mention. Each platform decides what it answers. flowgen removes the reasons a brand cannot be cited; it cannot force a citation.
No browser-level or user-level tracking of AI platforms. Measurements come from flowgen’s own sampled runs, not from users’ sessions.
No proof of cause from a before-and-after change. Performance shows the sequence of action and change so your team can judge, and uses untouched questions as a comparison where possible.
Platform answers change between runs. Scores are reported with their sample size, and trends matter more than single readings.
Coverage follows your plan. Check frequency and the platforms included are listed in each report, not assumed.

See the scores for your own site

One address is enough for a first reading of all six scores, with the questions behind them.

Run free GEO analysis Read the GEO Guide