Discover

A prompt library invented at a desk measures what you think customers ask. Discover builds one out of evidence: what your site says you sell, what people are actually searching for, and what they’re asking each other in public.

Two sub-tabs: From your site and Search trends.

From your site

An analysis of your own homepage — structure, content and messaging — usually about 30 seconds. It returns:

Section	What it gives you
Core offerings	What the site says you sell, in its own words
Customer segments	Personas the messaging targets, with an estimated split
Differentiators	The claims the site leans on
Key assumptions	What the analysis inferred rather than read — worth checking
Areas to validate	Where the positioning is vague or unsupported
Recommended prompts	Ready-to-add questions derived from the above

Two ways to read it.

As a mirror. The offerings and segments are what a language model concludes about you from a single visit — which is roughly what an AI engine does before deciding whether to recommend you. If they’re wrong, the answer isn’t more prompts; it’s clearer copy on the page.

As a prompt source. Recommended prompts come from the segments and offerings the analysis found, so they’re grounded in your actual positioning. Add the ones that match a question a customer would ask; skip the ones that read like marketing copy.

The keywords this analysis extracts also populate the keyword selector in Search trends.

Search trends

Pick a keyword, run the analysis, and get three things.

Interest over time

Search interest for that keyword over the past 12 months, alongside absolute volume figures: latest-month volume, a 12-month average, and a competition indicator.

The absolute numbers matter more than the growth percentage. A query can grow 1,000% and still be five searches a month — the panel flags those as Low volume rather than presenting them as opportunities, because relative growth on a tiny base is noise dressed as a trend.

Rising and top queries

Group	What it is	Use for
Rising queries	Fastest-growing related searches	Emerging demand you can be early on
Top queries	Highest-volume related searches	The head terms your category is decided on

Every query is converted into a probe-ready prompt, shown next to the raw signal it came from, so you can see exactly why a prompt is being suggested.

What real people are asking

Reddit discussions on the topic, with upvote counts, each refined into a prompt.

These are the most valuable prompts in the product, and the least like anything you’d write yourself: they carry the constraints, budgets and frustrations people only state to each other. They’re also self-reinforcing — community discussion is heavily retrieved by current AI engines, so a question asked on Reddit is often a question an engine has already read an answer to.

From signal to prompt

Everything in Discover ends the same way: Add to library. The prompt lands in the Prompts tab tagged with its source (Reddit, GEO, AI) and classified into an intent, ready to select for the next run.

1. Run the site analysis first It supplies the keywords the trends analysis needs, and tells you whether your positioning reads the way you think it does.
2. Pick keywords you actually compete on Not the widest term available. Trends on a term you can't win produce prompts you'll lose.
3. Take the community prompts Then the top queries, then the rising ones — filtering out anything flagged as low volume.
4. Check the intent mix afterwards Discover skews toward discovery prompts. Add comparison and trust prompts by hand or with the generator so the branded half of the picture isn't empty.

Prompts → What to do with the library once it's full. Content → Writing the page that answers a prompt you're losing.