AI prompt ideas generator
Put in a domain, get about 25 prompts its market plausibly asks an AI assistant — grouped by topic, split by intent.
A bare domain is enough. We read the homepage for context where the site lets us, then produce about 25 prompts for the market it sits in. Free, no sign-up, the whole list is shown.
These prompts are plausible, not measured. They are what a model says people in this market would ask an assistant. No prompt volume, assistant traffic or click data sits behind them: the assistant providers do not publish their query logs, and this tool consults no volume source of any kind. The way to find out which of these your market really asks is to track them.
Nothing generated yet
Put in the domain you want to be found for, or try one of these:
What this tool does, and what it refuses to claim
People do not type keywords into an assistant. They type sentences — “what is the best invoicing tool for a solo designer”, “is X or Y better for a small team” — and the assistant answers with a handful of named products. Whether you are one of those names is decided prompt by prompt, so the first practical question for anyone doing this work is simply: which prompts should I be watching? This tool produces a starting set.
It produces plausible prompts. It does not measure demand. The assistant providers hold their own query logs and publish none of them, so there is no public dataset of how often anybody types a given sentence into ChatGPT — and this tool consults none. We would rather hand you a couple of dozen well-shaped guesses and call them guesses than attach a number to each one that we could not stand behind.
The tool fetches the homepage of the domain you give it, reads the title, the meta description and the first few headings, and passes exactly that to one model under a fixed response schema. The result panel lists what was read, so you can see whether the model had anything useful to work with. If the fetch fails — a login wall, a bot block, a redirect loop — the page says so and falls back to the domain name alone rather than pretending otherwise.
The counts under the list are computed from the returned strings, not asserted: the informational/commercial split, the question-versus-comparison mix and the word-length band all describe the list you can see. They describe the list. They do not describe your market.
How to turn this list into something worth acting on
- Cut the ones you do not want to win. A generated list is generous by design. A prompt whose answer you could never credibly be part of is noise in every report that follows, and the cheapest time to remove it is now.
- Keep the commercial ones even when they hurt. “Alternatives to [you]” and “[you] vs [competitor]” are asked by people who already know your name and are deciding against you. They are uncomfortable and they are the highest-value prompts on the list.
- Add the five you know we missed. You know how your customers phrase things, including the internal jargon that never appears on your homepage. A model reading a marketing page cannot know that.
- Track them before you write anything. Writing a page for a prompt you have never checked is guessing twice. Run the prompts first, see which answers already name you, and write for the gaps rather than for all 25.
- Re-run it after a repositioning. This list is derived from what your homepage currently says. Change the positioning and the plausible prompts change with it — which is itself a useful check on whether the new copy reads the way you intended.
Why a prompt list is the first artefact, not the last
Almost every measurement in AI search is downstream of a prompt list. Mention rate is a rate across prompts. Share of voice compares you to competitors on the same prompts. A citation report is a list of which pages were cited in answers to prompts. Choose the list badly and every number built on it is precise and irrelevant.
That is also why the list is worth arguing about with someone who sells your product. Fifteen minutes of a salesperson’s attention on this page will beat any generator, including this one, because they have heard the actual sentences. Treat the output here as a first draft that saves you the blank page, not as a finished research artefact.
Questions people ask about this generator
How many people actually ask these prompts?
We do not know. Assistant providers do not publish their query logs, so we have no volume figure to attach to any prompt here — which is exactly why this tool shows none. Search-keyword volume answers a different question about a different surface; pasting it beside a prompt would make the number look measured when nothing here was measured.
Does this tell me whether my site appears in answers to these prompts?
No. This tool takes no measurement of any kind — it never asks an assistant anything about you. Finding out who gets named means running each prompt against the assistants repeatedly and counting, which is what prompt tracking in the product does on a schedule.
Why do most of the prompts not mention my company name?
Because the model is asked for them that way, on purpose. Unbranded prompts are the ones where an assistant picks the names it mentions, so they are where there is something to win; a list made only of “is [you] any good” covers people who already found you and would flatter you without teaching you anything. Branded and comparison prompts matter too, which is why some are asked for as well.
Will I get the same list twice?
No. The model call is non-deterministic, so wording and grouping move between runs. Treat a topic that shows up across several runs as the stable signal, and treat any single phrasing as one sample — one run is one sample, not a finding. If a prompt matters to you, copy it out; do not rely on it reappearing.
What do informational and commercial mean here?
Informational prompts are asked to understand something; commercial prompts are asked while choosing what to buy. The split matters because they need different pages: an explainer answers the first, a comparison or pricing page answers the second, and a site that only has one kind disappears from half the answers in its own category.
Do you store the domains I put in?
Not against you. The domain reaches our marketing service as a request and appears in its logs, and your address is counted against a per-tool rate limit so one caller cannot run this all day — but there is no account, no email field and no history you could be looked up in. The list is held for about a day in a shared cache keyed by the domain, which is what keeps a domain costing one model call rather than one per visitor. So the shareable link may return the cached list rather than generating a new one.
Where to go next
- Track these prompts across ChatGPT, Claude and PerplexityThis page suggests prompts. Prompt tracking runs them on a schedule and records which brands each answer names.
- Group your prompts into topics you can report onTwenty-five prompts is a list. Topics are how a list becomes something you can show a board and act on.
- See how one prompt fans out before anything is retrievedTake any prompt from the list above and watch a model decompose it into the sub-questions it would retrieve for.
- Check whether assistants recognise your brand at allIf the entity behind your domain is unclear, branded prompts are the first place that shows up.
- Read how mention rate is actually calculatedThe number every prompt list eventually rolls up into, including the sampling that makes it trustworthy.
- Compare your coverage against named competitorsThe comparison prompts above are where competitors get named instead of you — this is how you watch that happen.