Skip to main content Scroll Top

Free AI Visibility Checker: The Prompt Set You Run Yourself

Updated September 2026 · Written and maintained by the Progression Agency strategy team

A free AI visibility checker that does not pretend to query the assistants for you. It builds the twelve-prompt set, tells you how to run it, and gives you the sheet to record the answers in.

On this page · 11 sections
  1. Why most tracked prompt sets measure the wrong thing
  2. What this set does differently
  3. How to run it so the results mean something
  4. What these numbers cannot be turned into
  5. Reading the series rather than the month
  6. What we would not sell alongside this
  7. Adapting the set to your own sector
  8. What this AI visibility checker does and does not do
  9. The reference tables, in one place
  10. Everything else we have written on search, AI and getting found
  11. Using this as an AI visibility tracker rather than a one-off check

The short answerMost tracked prompt sets are mostly branded and mostly commercial, which produces a stable number that moves for reasons unrelated to your work. This builds a set weighted the other way: two problem-aware prompts that never name the category, two solution-aware, two informational, two commercial, two comparison and two brand. Run each in two assistants monthly, in fresh sessions, with the wording frozen, and record the answers verbatim with the date. That series is the only honest evidence available in this field.

No share-of-voice percentage, no rank and no reproducible score is available from this medium, and the page says so rather than inventing one.

AI visibility prompt set builder

Generates a twelve-prompt monthly set spread across the intents that matter, plus a recording sheet. The point is the ratio: ten prompts that are not about your name, and two that are.

Results appear here.

Runs entirely in your browser. Nothing you paste is uploaded, logged or stored — which is also why it takes pasted text rather than a URL: a browser cannot read another site’s files without that site’s permission.

Why most tracked prompt sets measure the wrong thing

Three biases recur in the first set anybody builds, and each of them produces a number that moves for reasons unrelated to the work.

They are mostly about the company’s own name

The first set anybody builds is six variations of their brand name, three product questions and a competitor comparison. It produces a stable, reassuring number that moves when somebody writes about you and stays flat otherwise. It does not tell you whether anyone who does not already know your name would encounter you.

They are mostly commercial

The second bias is toward buying-stage phrasing — best, top, who should I hire. Those are the questions most likely to be answered from directories and review platforms, which means the set is largely measuring somebody else’s property and reporting it as your score.

And the wording changes between runs

The most common invalidation. A prompt reworded slightly between months produces a different answer for reasons that have nothing to do with anything you did, and the series stops meaning anything. Whatever set you use, freeze the wording for its whole life.

The set most people build, and the set that tells you somethingThe set most people build, and the set that tells you something
A set dominated by your own name measures your name. It will look stable, it will move when somebody writes about you, and it will tell you nothing about whether the work is landing.

What this set does differently

Four deliberate choices, starting with the ratio between prompts about your name and prompts about everything else.

Two prompts in twelve are about your name

Enough to catch an inaccurate description of your business, which is a real and common problem, and not enough to dominate the result. The other ten test whether you appear to someone who has a problem rather than a supplier in mind.

It starts before the category word

The first two prompts describe a problem without naming the category. That is the widest and least contested entry point in almost every sector, and it is the one nobody tracks. If you appear there, you are genuinely visible; if you only appear once the category is named, you are visible to people already halfway to a decision.

It separates winnable from owned

Definition questions and who-is-best questions are included deliberately even though most businesses cannot win them. Tracking them tells you who does own them, which is useful context — and it stops anybody mistaking their absence for a failure of the work.

It asks the uncomfortable question

The twelfth prompt asks whether you are any good. That answer is being generated whether you look at it or not, and reading it monthly is the single most informative thing on the list.

Constant — Wording. Or you measure the wording.
Spread — Across intents. Not just commercial.
Verbatim — Answers recorded. Not summarised.
Dated — Every entry. Answers drift.
Two — Assistants. One is an anecdote.
Monthly — Forever. Boring is the point.

How to run it so the results mean something

Four conditions have to hold, and breaking any one of them quietly invalidates the series rather than obviously breaking it.

Fresh sessions, every time

Personalisation and conversation history both change answers. Start a new session for each prompt, or at minimum for each run, so that what you record reflects a cold question rather than a primed one.

Two assistants, not one

A single assistant’s answer is an observation. Two that agree is a weak signal; two that disagree is itself a finding, and usually a more interesting one. Use the same two every month.

Record verbatim, not as a score

The value is in the text: which sources were leaned on, how you were described, which competitor was named first. A yes-or-no column throws all of that away and leaves you with a number nobody can act on.

Record two metrics, not one

Whether you were named, and whether what was said was accurate. In several sectors an inaccurate mention is worse than an absence, and a set that only counts mentions cannot see the difference.

What a measurement set has to surviveWhat a measurement set has to survive
The last one is the real constraint. An elaborate measurement programme that stops after two months is worth less than twelve prompts run every month for a year.

What these numbers cannot be turned into

Three metrics this medium does not produce, and the one thing it does produce that is worth having.

Not a share of voice

There is no denominator. Twelve prompts in two assistants is a sample of a space with no known size, and dividing mentions by prompts produces a percentage that looks like a metric and is not one.

Not a rank

Nothing is ranked. An answer either draws on you or does not, and the order of names inside a sentence is not a position in a list.

Not a reproducible score

The same prompt in the same assistant on the same day can produce different answers. That is the medium, not a flaw in the method, and any provider reporting a precise score without a stated variance is reporting confidence they cannot have.

What it is, is a series

Twelve prompts, two assistants, recorded verbatim with dates, every month for a year. That is the honest deliverable, and after six months it is genuinely informative. What LLM visibility measurement can and cannot do sets out the reasoning in full.

Problem — No category word. Widest entry point.
Solution — Category level. Consideration set.
Informational — Definitions and process. Often publisher-owned.
Commercial — Cost and selection. Highest intent.
Comparison — Named where possible. Sharpest signal.
Brand — Identity and reputation. Accuracy, not volume.

Reading the series rather than the month

A single month tells you almost nothing; the pattern across intent bands over several months tells you a great deal.

A single month tells you almost nothing

Answers vary between identical runs, so a mention that appears in March and not in April is not evidence of anything. What is evidence is a direction sustained across several months, and a set that is only three months old cannot show one.

Look at intent bands, not the total

Twelve mentions out of twenty-four answers is a number with no meaning. Two of two problem-aware prompts naming you, where none did six months ago, is a finding. Read the set by band and the pattern becomes legible.

Pay attention to the sources, not just the names

The most actionable column in the sheet is usually which sources an answer leaned on. When the same directory keeps appearing, that directory is the surface for that question, and the work is your record there rather than another page on your own site.

What a tracked answer can and cannot tell youWhat a tracked answer can and cannot tell you
Four real outputs and four that are not available. Anyone reporting the bottom four with confidence is reporting a number the medium does not produce.

What we would not sell alongside this

Three products that would be easy to build on top of this data and would misrepresent what the data supports.

A dashboard with a single visibility number

It would be easy to build, it would look impressive, and it would be a composite of observations that do not support being composited. We would rather hand over twenty-four recorded answers than one confident number derived from them.

A guarantee tied to the set

Nothing about these prompts is under anybody’s control, so no promise about their output is honest. What can be committed to is the work that plausibly moves them and an accurate record of what happened.

A larger set as an upsell

More prompts produce more data and not more insight, and every set we have seen above about twenty stops being run by month four. If a bigger set is genuinely warranted, it is because you serve several distinct areas or categories, and the honest version is several small sets rather than one large one.

Named — Or not. The first metric.
Accurate — Or not. The second metric.
Competitors — Who appeared. Context.
Sources — What was cited. Where to work.
Date — Always. The series matters.
Verbatim — The text. The actual evidence.

Adapting the set to your own sector

Three edits worth making before the series starts, and one change that would undermine the whole design.

Replace the category word with what customers say

The generated prompts use whatever you typed, so the single highest-value edit is making that phrase the one your customers use rather than the one your industry uses. A practice that types ‘dental implantology’ will track a question nobody asks; one that types ‘dental implants’ will not.

Swap in the three things you actually get asked about

The service fields drive six of the twelve prompts. Use the three enquiries that arrive most often, described the way the caller describes them, rather than the three you would most like to sell.

Keep the intent spread even when you edit

Editing the wording is fine before the series starts. Dropping the problem-aware pair because they feel indirect is not, because they are the ones testing the part of the funnel everything else assumes you already own.

What each prompt type is actually worthWhat each prompt type is actually worth
Brand prompts sit at the bottom not because they are useless but because they are the ones everyone over-weights.

Want this done for your site?We build and maintain the search, content and paid programmes described on this page.

Get a free proposal

What this AI visibility checker does and does not do

It is worth being exact about this before you use it, because most things sold as an AI visibility checker work differently and cost money.

It does not query the assistants for you

There is no model call behind this page. It builds the set, explains how to run it and gives you the recording sheet; you run the twelve prompts yourself in two assistants. That is a deliberate design choice rather than a limitation we are apologising for.

Which is why it is free, and why the results are yours

A tool that queries models for you needs API keys, carries per-check cost, and hands you a number you cannot reproduce. Running the prompts yourself costs nothing, takes about twenty minutes a month, and produces answers you have read rather than a score you have been given.

What a paid AI visibility checker adds

Scale and automation: larger prompt sets, scheduled runs, dashboards, and multi-market coverage. If you are tracking many brands or many markets that is worth paying for, and we compare the options on our page about the tools. For a single business, the twelve prompts that matter are the twelve you would have chosen anyway.

How to check AI visibility without any tool at all

Open two assistants in fresh sessions, ask the twelve prompts this page generates, and write down what came back with the date. That is the whole method, it is what every honest paid tool automates, and nothing about it requires our page beyond the prompt set itself.

Share of voice — Not available. The medium does not produce it.
Rank position — Not available. There is no ranking.
One assistant — Anecdote. Use two.
Changed wording — Invalid series. Hold it constant.
Mostly branded — Measures your name. Not your visibility.
Summarised answers — Loses the finding. Record verbatim.

The reference tables, in one place

Four tables follow: the intent mix and why each is in the set, the recording columns, the mistakes that invalidate a series, and how to read a result. They summarise the sections above and are meant to be usable on their own.

The intent mix, and why each type is included
IntentPromptsWhy it is in the setWinnable?
Problem-aware2The widest, least contested entry pointUsually
Solution-aware2Tests presence in the consideration setOften
Informational2Definition and process; tells you who owns themPartly
Commercial2Highest intent; cost is usually shapeablePartly
Comparison2Sharpest signal, especially when namedPartly
Brand2Accuracy of description, not volume of mentionsYes
Total12Ten not about your name, two that areMixed by design
What to record for every answer
ColumnWhy it is thereCommon mistake
DateAnswers change without noticeOmitted, making the series unusable
AssistantTwo behave differentlyMixing them in one column
IntentLets you read the pattern by typeDropped, leaving twelve unrelated rows
PromptProves the wording did not changeParaphrased between months
Were you namedThe first metricTreated as the only metric
Named competitorsShows who occupies the answerIgnored
Anything inaccurateThe second metricNot captured at all
Answer verbatimWhere every real finding livesReplaced with a score
What invalidates a series
MistakeEffectFix
Changing the wordingThe series measures the wordingFreeze it for the whole year
Using one assistantAn anecdote, not a signalUse two, the same two
Carrying session historyAnswers are primedFresh session each run
Recording a score instead of the textFindings are lostRecord verbatim
Skipping monthsTrends become noiseSame day each month
Mostly branded promptsMeasures your nameTen of twelve should not be
Reporting a percentageImplies a denominator that does not existReport counts and text
How to read a result
ObservationWhat it suggestsWhat to do
Named on problem-aware promptsGenuine visibilityKeep doing what produced it
Named only on brand promptsVisible to people who already know youWork the earlier intents
Named but described inaccuratelyAn entity problem, not a content oneFix the records off your site
Absent on definition promptsNormal; publishers own themDo not chase it
A competitor named consistentlyThey occupy that answerRead what they publish on it
Answers cite a directoryThe directory is the surfaceFix your record there
Two assistants disagree sharplyThin or conflicting sourcesPublish the clearer version
Prompt types by what they revealPrompt types by what they reveal
The bottom-left is where your own work shows up. The top-right belongs to publishers and directories and should be tracked for context, not for progress.
The monthly routineThe monthly routine
Twenty-four answers a month. It is dull, it takes twenty minutes, and after six months it is the only honest evidence anybody in this field has.
The set, in numbersThe set, in numbers
Two of twelve. If your current set is mostly branded, that is the first thing to change.
Fill — Five fields. Two are optional.
Generate — Twelve prompts. Across six intents.
Download — The CSV sheet. Twenty-four rows.
Run — Same day monthly. Fresh sessions.
Record — Verbatim and dated. No scores.
Review — Quarterly. Trends, not months.
Ratio — Ten to two. Not the reverse.
Wording — Frozen. For the whole series.
Fresh — Sessions. No carried context.
Both — Named and accurate. Two metrics.
Honest — About limits. No invented scores.
Sustained — Twelve months. The only real evidence.

Everything else we have written on search, AI and getting found

Using this as an AI visibility tracker rather than a one-off check

A single reading tells you very little, because assistant answers are not deterministic. The same prompt can return a different list an hour later.

What is worth tracking is the direction over months, measured with identical wording. That is what the prompt set and the CSV are for: they turn an impression into a series.

What makes a tracker rather than a checker

  • A fixed prompt set, saved and re-used verbatim rather than rewritten each time.
  • A recorded date for every reading, because an answer is only true for the day it was given.
  • The same assistants each round, since results differ meaningfully between them.
  • A record of which competitors appeared, not only whether you did.
  • A cadence slow enough that you are watching signal rather than noise.

Why identical wording matters more than it sounds

Rephrasing a prompt changes the answer. If you rewrite the question between rounds, you are measuring your own rewriting rather than your visibility. The CSV exists so the wording is fixed in a file rather than reconstructed from memory each month.

A workable cadence

  1. Build the prompt set once and save the CSV.
  2. Run every prompt in at least two assistants and record the date.
  3. Note whether you were named, in what position, and whether you were linked.
  4. Record which competitors appeared, because that is the more actionable half.
  5. Repeat monthly. Treat a single-mention change as noise, not a trend.
  6. Review the direction quarterly against what you actually changed.

What tracking cannot give you

It cannot give you a share-of-voice figure across all conversations, because nobody observes all conversations. Any vendor presenting a total share number is extrapolating from their own sample. What you can honestly hold is your own series, measured the same way each time.

Checking against tracking
A one-off checkTracking over time
Tells youWhether you appeared todayWhether you are appearing more or less
Prompt wordingWhatever you typedFixed and re-used verbatim
Number of readingsOneOne per period, recorded
Useful forA first lookDeciding whether work is paying off
Main riskMistaking one reading for a verdictChanging the wording between rounds
Honest outputA snapshotA direction, not a share of voice

AI, AEO and what is changing

Websites and design

Choosing and working with an agency

Social, content and brand

By industry and by situation

Want this run for you, every month

We run the set, record the answers verbatim, and report what changed and why — including the months where nothing did.

/contact

Not sure which of these applies to you?Tell us the situation and we will say plainly what we would do first, and what we would not.

Talk it through

Getting found in search

AI, AEO and what is changing

Paid media and lead generation

Websites and design

Choosing and working with an agency

Social, content and brand

By industry and by situation

Frequently asked questions

Is this a free AI visibility checker?
Yes, with one honest qualification: it does not query the assistants for you. It builds the twelve-prompt set, explains how to run it and gives you the recording sheet — you run the prompts yourself, which is why it is free and why the answers are ones you have read rather than a score you have been handed.
How do I check my AI visibility?
Open two assistants in fresh sessions, ask a set of twelve prompts spread across six intent types, and record the answers verbatim with the date. Repeat monthly with the wording unchanged. That is the whole method; this page generates the set for your business.
What does a paid AI visibility checker do that this does not?
Scale and automation — larger prompt sets, scheduled runs, dashboards and multi-market coverage. Worth paying for if you track many brands or markets. For a single business the twelve prompts that matter are the twelve you would have chosen anyway.
Can an AI visibility check give me a percentage?
No, and we would be sceptical of one that does. Twelve prompts in two assistants samples a space of unknown size, so there is no denominator and any percentage is a number dressed as a metric.
How do I track AI visibility over time?
Fix a prompt set and re-use the wording verbatim, run it in at least two assistants, record the date every time, and note which competitors appeared alongside you. Monthly is a sensible cadence; compare direction rather than individual readings.
Why does the wording have to stay identical?
Because rephrasing a prompt changes the answer. If the wording moves between rounds you are measuring your own rewriting rather than your visibility.
How often should I run an AI visibility tracker?
Monthly for most businesses. Weekly produces movement you cannot act on, and a change of a single mention is noise rather than a trend.
Can a tracker tell me my share of AI answers?
No, and no product can. Nobody observes every conversation, so any total share-of-voice figure is extrapolated from a private sample. What you can honestly hold is your own series, measured identically each time.
What should I record each round?
The date, the assistant used, whether you were named, your position among named businesses, whether you were given a link, and which competitors appeared. The competitor column is usually the more actionable half.
What does this tool produce?
Twelve prompts spread across six intent types, tailored to your business, plus a CSV recording sheet with twenty-four rows — each prompt, in each of two assistants.
Why only two brand prompts out of twelve?
Because a set dominated by your own name measures your name. It produces a stable number that moves when somebody writes about you and tells you nothing about whether someone who does not know you would encounter you.
What is a problem-aware prompt?
One that describes the problem without naming the category — ‘what should I do about a burst pipe’ rather than ’emergency plumber’. It is the widest and least contested entry point in most sectors, and almost nobody tracks it.
Why include questions I cannot win?
Because knowing who does own the definition and who-is-best questions is useful context, and because it stops anybody mistaking your absence there for a failure of the work.
Why does the wording have to stay the same?
Because a reworded prompt produces a different answer for reasons unrelated to anything you did. Changing the wording mid-series is the most common way a tracking programme quietly becomes meaningless.
Why two assistants?
One answer is an observation. Two that agree is a weak signal, and two that disagree is itself a finding — usually a more interesting one than either answer alone.
Why fresh sessions?
Conversation history and personalisation both change answers. A primed session tells you what the assistant will say after you have already told it about yourself, which is not the question.
Should I record a score or the answer?
The answer, verbatim. The value is in which sources were leaned on, how you were described and which competitor was named first. A score throws all of that away.
What two metrics should I track?
Whether you were named, and whether what was said was accurate. In several sectors an inaccurate mention is worse than an absence, and a set that only counts mentions cannot tell the difference.
Can I turn this into a share-of-voice percentage?
No, and we would be sceptical of anyone who does. There is no denominator — twelve prompts in two assistants samples a space of unknown size, so a percentage looks like a metric without being one.
Is there a rank position to track?
No. Nothing is ranked. An answer either draws on you or it does not, and the order of names inside a sentence is not a list position.
Why do answers change between identical runs?
That is the medium. The same prompt in the same assistant on the same day can produce different text, which is why a single run is an observation and a series is evidence.
How long before the series is useful?
About six months. Before that you have observations; after it you have a trend you can reason about, including the months where nothing moved.
How long does a monthly run take?
Around twenty minutes for twenty-four answers. It is dull, and the dullness is the point — an elaborate programme that stops after two months is worth less than twelve prompts run for a year.
Should I add more prompts?
Only if you will still run them in month nine. Coverage is worth less than consistency, and twelve is about the limit of what most teams sustain.
What if I serve several areas?
Run the location-sensitive prompts once per area rather than expanding the whole set. Location changes answers more than almost any other variable.
Should I include my competitor’s name?
Yes, if you have one clear comparison. A named comparison is answered on stated facts, which makes it the sharpest signal in the set and a direct prompt to publish your own version of those facts.
What does it mean if answers keep citing a directory?
That the directory is the surface for that question. The fix is your record on that directory, not another page on your site.
Does the tool send my inputs anywhere?
No. Everything runs in your browser and the CSV is generated locally; your inputs are never uploaded. The only thing sent to us is the contact detail you enter to unlock the full findings.
Can I use this for clients?
Yes, with no restriction. The recording sheet is deliberately plain so it can go straight into a spreadsheet or a report.
What should I do if I am named but described wrongly?
Treat it as an entity problem rather than a content one. The description is usually assembled from records off your own site — directories, profiles, listings — and that is where the fix is.
Is this the same as an AI visibility tool subscription?
No. Paid tools run larger prompt sets automatically and produce dashboards. This produces the same underlying evidence by hand, honestly labelled, for nothing — and for most businesses the twelve prompts that matter are the twelve they would have chosen anyway.

Want this done for your site?We build and maintain the search, content and paid programmes described on this page.

Get a free proposal

Get a free marketing proposal

Tell us what you are trying to grow and we will come back with a plan, not a pitch deck. Same-day reply on weekdays.

Our website uses cookies from third party services to improve your browsing experience. Read more about this and how you can control cookies by clicking "Privacy Preferences".
Privacy Preferences
When you visit our website, it may store information through your browser from specific services, usually in form of cookies. Here you can change your privacy preferences. Please note that blocking some types of cookies may impact your experience on our website and the services we offer.
Privacy Policy
You have read and agreed to our privacy policy
Required
Contact Us