Your prompt is the most important setting on a Radar, because it is used twice: once to generate the marketplace searches, and again to score every listing those searches return. A vague prompt produces vague searches and a lenient scorer, so it fails at both ends. The fix is almost always to state your limits explicitly rather than your preferences.
How your prompt is used
When you save a Radar, the AI reads your prompt and derives the specific searches to run — that is what you confirm on the Review step. Those searches go out to each marketplace you have enabled.
Everything they return then comes back to the same prompt. The AI compares each listing’s title, description, price, and attributes against your criteria and scores how well it fits. Listings that score well enough become Pings; the rest are discarded.
This is why the prompt rewards precision on both fronts. Words that make a good search term are not always the words that make a good filter, and a prompt that does both jobs well tends to name the item plainly and then be specific about the boundaries.
What to include
Aim for two or three sentences covering:
- The item. Be as specific as you genuinely are. “Toyota Tacoma” beats “pickup truck” if that is what you actually want; “road bike” beats “bike” if you would not drive across town for a mountain bike.
- Your price range. Both ends. A maximum stops the AI surfacing things you cannot afford; a minimum quietly filters out the broken and the miscategorised.
- The conditions that matter. Mileage, size, condition, model year, material — whatever would make you say no.
- Your deal-breakers, stated as exclusions. “No salvage titles.” “Not interested in project bikes.” “No glass tops.”
What does not work
- Facts that are not in a listing. The AI only sees what the seller wrote. It cannot know whether a car has been garage-kept, whether the seller is easy to deal with, or whether an item is genuinely a rare variant if the listing does not say so.
- Preferences with no limit. “Cheap”, “good deal”, and “nice condition” give the scorer nothing to measure. Attach a number or a concrete standard.
- Several unrelated items in one prompt. A prompt covering couches and power tools and bikes produces searches that are diluted for all three. Use a separate Radar per category.
- Keyword stuffing. Repeating “truck pickup 4x4 vehicle automobile” does not broaden the search — it just makes your criteria harder to score against.
Examples
Too vague
Looking for cheap furniture
Nothing here bounds a price, a type, or a condition. The searches will be generic and almost everything will score as a partial match.
Better
Solid wood dining tables and dressers under $250, in good or better condition. Mid-century style preferred. No particle board, no glass tops.
Too vague
Trucks near me
Location is already handled by your territory, so it adds nothing, and “trucks” spans a $2,000 beater and a $60,000 diesel.
Better
Toyota Tacomas or Tundras between $4,000 and $20,000, under 200,000 miles. Manual transmission preferred. No salvage or rebuilt titles.
Too vague
Good deals on tools
Better
DeWalt or Milwaukee cordless power tools under $150 — drills, impact drivers, and combo kits. Batteries and charger should be included. Not interested in corded tools or bare tools sold on their own.
Tuning a prompt that is not working
Change one thing at a time, and give it a day between changes — criteria edits are rate limited to roughly one per 24 hours anyway. See creating a Radar.
Getting nothing? Your prompt is probably narrower than the market. Widen the price range, drop the least important condition, or add an adjacent item type. Check the Review step’s searches too — if they came out oddly specific, your phrasing was ambiguous.
Getting too much? Add the limit you left implicit. Convert “preferred” into a requirement, tighten the price ceiling, and write your rejections in explicitly. If the listings are technically on-target but not worth your time, the faster fix is the notification level rather than the prompt — see notifications.
Getting the wrong category entirely? Look at the derived searches. If the AI misread which item you meant, naming the category outright in your first sentence usually fixes it.
Related: Creating a Radar · Notifications · Pings and listings
Frequently asked questions
How long should a Radar prompt be?
Between 10 and 500 characters. Two or three sentences is usually right — enough to name the item, your price range, and your deal-breakers, without padding it with detail the listing will not contain.
Should I write my prompt like a search query or a sentence?
A sentence. The prompt is read by an AI, not matched as keywords, so plain English describing what you want and what you would reject works better than a list of search terms.
Why is my Radar returning listings that do not match?
Usually the prompt states a preference without stating the limit. "Cheap dining tables" gives the AI nothing to measure against; "dining tables under $200, solid wood, no glass tops" does.
Can I tell my Radar what to exclude?
Yes. Write exclusions into the prompt in plain English — "no salvage titles", "not looking for project bikes" — and the AI applies them when scoring each listing.