How GEOflux works

Where every number comes from, why AI answers change from day to day, and the words used across GEOflux

Every number in GEOflux comes from one loop: you choose the questions, GEOflux asks them, reads the answers, and counts. Knowing that loop is what lets you trust a number, and spot when one is not telling you what you think.

The loop

You define the prompts. A prompt is a question a buyer would ask an AI assistant, such as "Which CRM should a 10-person agency use if we mostly need pipeline tracking?". You can attach personas to a prompt to ask it again on behalf of a specific audience.
GEOflux asks every engine. Each prompt goes to ChatGPT, Gemini and Perplexity through the same web interfaces people use, so the answer is the one a real person would get, including whether the engine searched the web first. A new prompt is asked within about a minute of being saved; after that it is asked every day on Starter, Growth and trials, and every 3 days on Free.
The full answer is stored. You can read it, search it and export it on the Answers page, exactly as the engine returned it.
The answer is read. GEOflux finds every brand the answer names (yours, your competitors, and any other brand), scores the tone toward each one from 0 to 10, and collects the websites the answer cites.
The results are counted over your scope. The Overview and every other page add those readings up over the period and filters you pick. Only completed answers count; an answer that is still being collected or failed is left out.

The same loop runs for every project. A project is one brand in one country, with its own prompts, competitors, plan and credits. GEOflux has no language setting: the answers come back in the language you write your prompts in.

What counts as a mention

An answer mentions a brand when it names it at least once. Naming a brand three times in one answer still counts as one mention, so mentions are always a count of answers.

GEOflux recognizes your brand by its name and the alternate names you add on My brand, such as abbreviations and other spellings. A product name that includes your brand name, such as "Wise Business" for Wise, also counts as your brand when the answer ties it to your website, without an alternate name. Competitors are recognized the same way, from the names you give them on the Competitors page.

If AI talks about you under a name GEOflux does not know, those answers read as not mentioning you. Adding alternate names is the cheapest accuracy fix there is.

Why answers change from day to day

AI answers are not fixed. Ask the same question twice on the same day and you will rarely get the identical list of brands. Over a month even a category leader is missing from some answers, while smaller brands come and go. Engine updates, the user's location and small changes in wording all move the answer too.

That is the medium, not a fault in your setup. It is why GEOflux asks every prompt on a schedule and reports shares across many prompts and days rather than the position in any one answer:

Read shares, not single answers. A visibility of 40% over 30 days and 50 prompts is a solid number. Whether you appeared in one answer on Tuesday is not.
Read trends over weeks. A 14-day read shows where you are absent; a 30-day read is a baseline you can compare against.
Change one thing at a time. When a number surprises you, change one filter and see which one moves it.

See Build a prompt plan for how to set a baseline, and When data updates for how today's partial numbers behave.

Every number has a scope

A visibility score means nothing without its scope: the period, the engines, the topics, the Buyer Journey stage, the personas. The same brand can read 70% on Perplexity for one topic and 15% on ChatGPT across everything. The filter bar at the top of every page sets that scope, and the number on screen is always computed for it. See Filters and time periods.

Pick one headline scope and keep it, for example last 30 days, all engines, non-branded prompts. Comparing two numbers taken over different scopes is the most common way to draw the wrong conclusion.

Words used across GEOflux

Term	What it means
Project	One brand in one country, with its own prompts, competitors, plan and credits. Agencies hold one project per client brand.
Prompt	A question GEOflux asks on your behalf. At least 4 words and at most 200 characters.
Answer	One engine's response to one prompt, for one persona or none, on one run.
Mention	An answer that names a brand at least once.
Citation	A link an answer gives as a source.
Competitor	A brand you track so your numbers have something to be compared with.
Persona	A described audience, added to the prompt as "about me" context. It runs as a separate answer next to the prompt's own. See Personas.
Topic	The one theme a prompt belongs to, such as "Espresso machines". Your main reporting unit.
Buyer Journey	Where the buyer is: Discover, Evaluate or Buy. One per prompt.
Brand type	Branded (the prompt names your brand or a competitor) or Non-branded (it names no brand). One per prompt.
Tag	A free label for anything else you want to filter on, such as a campaign. Up to 10 per prompt.
Credit	One AI answer. A tracked prompt holds 1 per engine that answers it (3 on ChatGPT, Gemini and Perplexity), and as many again per persona attached to it. Credits are held while the prompt is active and returned when you archive it. See Billing and credits.
Scope	The period and filters a number is computed over.

For how topics, tags, Buyer Journey and Brand type work together, see Organize your prompts.