Research your prompts

Prompt research gives you the prompts your customers would really ask about your business, ready to track on ChatGPT and Gemini, and keeps the ones you like in a library for your brand. Nothing here costs anything until you choose to track a prompt.

Where it lives

In the sidebar, open Research > Prompt Research (/dashboard/prompts), then choose a location. If the location is one branch of a bigger brand, the library you see is the brand's, and the branch you pick decides whose evidence you are reading and where a tracked prompt is asked.

How suggestions are written

There is no such thing as prompt search volume. Nobody publishes it, nobody can measure it, and any tool showing you a number for how many people asked ChatGPT a particular question is inventing it.

So Prompt research starts from what your own data shows. For each branch, our agent reads the keywords you track, the searches that already find you on Google and Maps, the searches ChatGPT and Gemini ran about you, what your reviews keep raising, what your Google listing and website say, and anything you've added yourself. It writes a short set of prompts the way your customers ask, spread across what they're doing: ready to choose, describing a problem, needing someone now, having a must-have, looking for someone right for their situation, comparing options, watching the cost, planning for an occasion, or early research. Anything that does not fit your kind of business is left out.

Before you see a suggestion, we check it. Every prompt:

• names your area, and never says "near me" or another town
• names no business, yours or anyone else's
• asks for recommendations, because an answer with no names measures nothing
• is about something your evidence shows you offer
• isn't a repeat of a prompt you track, saved or dismissed, or the same sentence with a different service swapped in

The strip under the page title shows which sources are feeding the branch you're looking at. A dashed source isn't feeding it yet, and clicking one takes you to where you connect it.

Each suggestion has a chip saying what it was written from. Click it to see every piece of evidence behind the prompt:

Chip	Where it comes from
You track	A keyword you track on Google
You told us	A file you uploaded or pasted: enquiry forms, price list, FAQs
Your Google listing / Your website	What your listing or website says you offer
Search Console	Searches that already show your site on Google, with their impressions
Maps	What people typed into Google Maps to find your profile
You rank for this	A Google search your website already ranks for
A customer review	Something a customer wrote about you
Engines searched	A search ChatGPT or Gemini ran before answering about you, and how often
Your reviews	Something your customers keep writing about
Google asks / Related search	Shown by Google around a search you track

Suggestions are written the first time you open a branch, and kept. After you've added keywords or evidence, press Write fresh suggestions for a new set. It replaces the suggestions on screen, and anything you've saved stays saved. A branch can get a fresh set once every six hours.

Suggested, Saved, Tracked

Prompts move through three states, and only the last one costs anything. The three are the tabs above the list, in that order, with what your tracked prompts cost each month on the last one:

• Suggested. Written for this branch from its evidence. Press Track to start measuring one straight away, Save to keep it for later, or the cross to dismiss it so it is never suggested again. You can edit the wording first.
• Saved. Your brand's library. Saved prompts wait until you want them, and Track here starts measuring one at this branch. A saved prompt written for one branch is asked about the right town when you track it at another.
• Tracked here. Asked on every run at the branch you are looking at. Reword one with the pencil or stop asking it with the cross.

To do several at once, tick them (or tick a need's heading to take all of its prompts) and use the bar at the bottom of the screen. It shows what they will cost in checks each month before you press Track.

If a prompt doesn't match any keyword you track at the branch, you're asked which keyword it should be measured under. It's saved either way, so choosing Leave it in Saved loses nothing.

Prompts that were already being tracked show where they came from too. A starter prompt was written from your business category before there was any evidence behind it, which is worth knowing when you decide what to keep.

One need, measured once

"dermatologist london", "dermatologist in london" and "london dermatologist" are three keywords and one thing your customers want. Prompt research groups keywords like these into one need, and they share one set of prompts instead of paying to ask the same question three times.

If we group two keywords that really are different, open the ... menu on the need's heading and choose the keyword that is a different need. It gets its own prompts from then on, and the same menu merges it back later.

The research map

Each slice of the map is a need, and it fills from the centre as you measure it: solid is tracked, the middle shade is saved, and the pale outer part is suggested. A slice with no solid part, and Not tracked under its name, is a need you have evidence for and are not measuring yet. The centre counts how many of your needs are tracked.

Click a slice, or a need in the list beside the map, and its panel opens with everything that need has at this branch:

• Tracked here: every prompt asked about it, grouped by stage (ready to buy, comparing, exploring) with situation prompts on their own. Click the pencil to reword one, or the cross to stop asking it. A prompt that has already run keeps its results with its old wording. A keyword always keeps at least one prompt, so the last one can be reworded but not removed.
• Suggested: track one, track them all, save one for later or dismiss it.
• Saved for later: track one here when you're ready.
• Write your own prompt: type it the way a customer would ask, then Track it or Save for later.

Every need on the board has an Open button that opens the same panel. On a phone you get the list without the wheel.

The same map and panel are on the location's Coverage setup page, reading the same prompts, so a change made on one page shows on the other.

Your own sources

The strongest evidence is the data only you have. Press Add your own sources under the page title. You are asked what kind it is first, because a price list and a parking policy speak to different customers:

• Questions people ask you. Enquiry or contact form exports, live chat logs, the questions reception answers all day, site search. These become prompts almost word for word.
• Services you offer. Service list, treatment menu, price list, and the jobs you take that your website barely mentions. These find needs nothing is measuring yet.
• Who you are. Practitioners and specialisms, accreditations, insurers accepted, languages spoken, whether you take new clients. These become situations to test, such as paying privately or taking new patients.
• The practical stuff. Opening hours, out-of-hours cover, parking and access, payment methods, home visits, wheelchair access. These become situations to test, such as out of hours or getting there.

Messy exports are fine

Export straight from your form plugin, inbox or CRM. You do not need to tidy the file first:

• We work out the delimiter, skip the header row, and pick the column that actually holds the text. The page tells you which column we used, so a wrong pick is obvious rather than mysterious.
• Each line is cleaned: HTML and bullets removed, smart quotes normalised, spreadsheet formula characters stripped, repeats and signatures dropped.
• Contact details are removed before anything is stored. Email addresses, phone numbers, links and long numbers are replaced with a label. The evidence is the question, never who asked it.
• Lines that read like an instruction to an AI are dropped rather than stored, because our agent reads this text later.

Then you see exactly what we would keep, with a note on every line we changed and a reason for every line we left out. Untick anything you do not want. Nothing is saved until you approve it.

Files must be CSV, TSV or plain text, up to about 400KB. Spreadsheets and documents are not read: export them, or paste the column you care about. Please don't upload card details or health records.

Next steps

• Decide how deeply each keyword is measured: press Coverage setup at the top of this page. To shape the prompts themselves, see Choose the prompts AI Engine Visibility runs.
• Add the situations your customers arrive with: Ask as the customers you actually get.
• See how you are doing on them: Track AI Engine Visibility.