Choose the topics and prompts to test
Six tools turn your website, your Search Console data and the searches AI platforms run into a set of topics and unbranded prompts, then test them on a schedule.
Start at app.dejan.ai
Every figure in the app depends on the questions you choose to test. Our public demo project tracks 63 entities, 43 of them taken from a Treewalker session, and holds 272 citation prompts, none typed by hand. Search Console showed that queries labelled NAVIGATIONAL brought 616 of 619 clicks, so the topic queries had almost no clicks (3 of 619). A scheduled run of 10 unbranded prompts named Algoroo in 4 answers and cited algoroo.com in none.
The workflow at a glance
| Step | Tool | Question it answers | Demo result |
|---|---|---|---|
| 1 | Treewalker | What do models associate with my site? | 81 items; 43 became entities |
| 2 | Organic Search | Which queries could win the most traffic? | 189 queries ranked by potential |
| 3 | Intent Classification | Which clicks are brand and which are topic? | NAVIGATIONAL: 616 of 619 clicks |
| 4 | Query Gaps | Which tracked queries never appear in Search Console? | 48 queries |
| 5 | Prompt Generator | How do I write prompts at scale without the brand name? | 20 prompts in one call |
| 6 | Citation Mining | What do the answers say on a schedule? | Named in 4 of 10, cited in 0 |
1. Treewalker: collect candidate topics
A Treewalker session lists what a model associates with your website from memory and with search. The Discovered Entities sub-tab holds every item from both sessions, and one click adds an item to your entity list. The two demo sessions listed 81 distinct items, and 43 of the 63 tracked entities came from it.
2. Organic Search: rank your queries by potential
Connect Search Console and import. Query Potential multiplies each query's impressions by its average position, so a large audience at a poor position scores highest. Algoroo's top query was "algorithm tracking tool google": 94 impressions at position 28.6 and 0 clicks.
3. Intent Classification: separate brand from topic
Label every query with an intent set or with your own entities. With the CLINT labels, the demo's 67 classified queries split into 41 NAVIGATIONAL, 37 INFORMATIONAL and 16 COMMERCIAL, with more than one label allowed. By clicks the split is 616 of 619 for NAVIGATIONAL, the brand and its misspellings. The INFORMATIONAL and COMMERCIAL queries are the topics to test.
4. Query Gaps: find tracked queries Search Console never recorded
Query Gaps compares the queries you track for AI visibility with your Search Console data and lists the ones whose exact wording never appears as a Search Console query. The demo lists 48, such as "current google search volatility" and "google algorithm update alerts". A close variant may still get traffic, so read each one as a wording to cover, not proof that no page of yours shows in Google.
5. Prompt Generator: write unbranded prompts
Give the Prompt Generator your project context, a page, a query or an entity, and it writes 10, 20, 50 or 100 prompts in one call. "Exclude branded terms" is ticked by default, so the prompts test whether the answers name you when the question does not name your brand. The demo library holds 272 prompts: 142 written from entities, 100 from queries, 20 from the Prompt Generator and 10 from a URL.
6. Citation Mining: test the prompts on a schedule
Save the prompts as a Prompt Set and schedule it Daily, Weekly or Monthly, so the same questions run over time. In scheduled run #13054, Google answered 10 prompts about tools that track algorithm updates. The prompts averaged 35.5 words and the searches Google ran averaged 8.5. 4 answers named Algoroo and none cited algoroo.com.
Open run #13054 · Citation Mining guide · Prompt to Query guide
Where this workflow leads
Run an AI visibility audit
Find where AI answers name and cite your competitors and not you.
Correct what AI models remember
Find what models get wrong about your brand.
Win the AI citation for one page
Move a page from named to cited.
Start your first project
Your first project includes a complimentary baseline run. No card required.