Do AI Visibility Tools Monitor Fixed Prompts or Generate Prompts Automatically?

How do AI visibility tools choose which prompts to track?

Tools use one or more of four approaches:

Most tools combine the first three: they suggest prompts, you accept or edit them, and the accepted list is tracked as a fixed set. Some rotate prompts automatically; with those, ask whether you can lock a set. Some examples: Squarespace's built-in AI Visibility tool suggests branded and non-branded prompts that you add or remove; Otterly's AI Keyword Research turns keywords into prompts; and keyword-based trackers such as LLMrefs generate the prompts for each keyword themselves and report results per keyword, so you freeze the keyword list rather than the prompts. With that last type, ask whether the generated prompts stay the same between periods.

Some tools also attach a "volume" estimate to each prompt. Treat these carefully; there is no public source of how often people ask AI engines a given question, and the numbers are modelled. See whether AI prompt volume data is reliable.

Why do trends need a fixed prompt set?

A visibility trend compares the same thing at two points in time. If the prompts change between those points, the trend mixes two effects: real change in how engines answer, and the change in what you asked. A rise from 30% to 45% can come entirely from swapping in prompts where you already do well.

That is why the core of any tracking should be:

• Frozen: the same prompts, word for word, every period.
• Versioned: when you do change the set, record the date and the change, and compare only like with like.
• Separate from exploration: new prompts go into a discovery group until they have enough history to join the core.

A fixed set also makes it harder to fool yourself or others with cherry-picked prompts, a problem covered in how to detect cherry-picked prompts in a case study.

What do generated prompts get wrong?

Generated prompts are useful, but check them before tracking. Common problems:

• Written from your point of view. Prompts generated from your own website tend to use your product language and your positioning, not the words buyers use. They can make you look more visible than you are.
• Branded when they should not be. Prompts that include your brand name measure whether engines know you, not whether they recommend you. Keep branded prompts in their own group.
• Missing conditions. Real buyers add budget, location, company size, integration or deadline. Generated prompts are often shorter and more generic, and generic prompts produce different answers.
• Near-duplicates. Several prompts asking the same thing in slightly different words inflate the count without adding coverage.
• Off-market. Prompts for customers you do not serve, or problems you do not solve, dilute the results.

A quick review rule: for each generated prompt, ask whether a real buyer would type something close to it before choosing a supplier like you. If not, drop or rewrite it.

Do you need prebuilt prompt sets or custom prompts?

Custom prompts are not strictly required; prebuilt and generated sets are there for convenience. But the prompts you track decide what your numbers mean, so for anything you report or act on, a reviewed, custom core set is worth the effort.

A practical split for most businesses:

Prebuilt or generated prompts are a good way to fill the discovery group and to spot questions you had not thought of. They are a poor substitute for the core group. How to write the core group is covered in how to choose the right prompts to track.

How can you tell whether a tool's prompt set represents your market?

Run a short audit on any set before you trust the numbers it produces:

• Match it against real questions. Collect ten questions buyers actually asked your team last month. Count how many have a close equivalent in the set. Fewer than half means the set misses your market.
• Count branded prompts. If more than a small share mention your brand, the headline visibility figure will look better than your real standing.
• Count conditions. Check what share of prompts include a budget, location, size, feature or deadline. If almost none do, the set is more generic than your buyers.
• Look for clusters of near-duplicates and merge them, so one topic does not dominate the average.

Twenty minutes on this check often changes which numbers you report.

What should you ask a tool vendor about prompts?

Before choosing a tool, ask:

1. Can I enter and lock my own prompts exactly as written?
2. Does the tool ever change tracked prompts automatically? If so, can I turn that off, and is every change logged?
3. How are suggested prompts generated, from what data, and can I see why each was suggested?
4. Can I group prompts such as core, branded, control and discovery, and report on each group separately?
5. How many runs per prompt per engine does the tool make, and on which schedule?
6. Where do prompt volume numbers come from, if the tool shows them?
7. Can I export prompts and answers, so the set can move with me if I switch tools?

A tool that cannot lock a prompt set, or changes it without telling you, cannot give you a trustworthy trend.

How do you combine fixed and generated prompts in practice?

A routine that works for most teams:

• Month one: build the core set from real buyer language; use the tool's suggestions to check for gaps. Run everything for four weeks before drawing conclusions.
• Every month: report trends from the core set only. Review the discovery group; if a discovery prompt consistently matters, move it into the core set from the start of the next period and note the change.
• Every quarter: review the whole core set against new sales and support questions. Retire prompts that no longer matter, but keep their history.

This keeps trends honest while still letting the set evolve with your market.

What do people ask about prompt sets in AI visibility tools?

How many prompts should a small business track?

Twenty to forty core prompts are usually enough to cover the main buyer questions. More prompts add cost and review time without always adding insight.

Should I use the prompts my tool suggests?

Use them as ideas. Review each one against real buyer language, remove branded and duplicate prompts from the core set, and add the conditions your buyers usually include.

Can I compare my results with a competitor's tool report?

Only if both use the same prompts, engines and run counts. Different prompt sets produce different numbers even for the same brand.

How often should the core prompt set change?

Rarely: once a quarter at most, with every change recorded. Frequent changes make month-to-month trends meaningless.