Understand AI visibility tracking in Vestis Atlas
Vestis Labs
Last Update 2 months ago
A simple guide to checking whether AI answer engines mention your brand and products - and how to read the results.
AI visibility tracking is a repeatable test, not a placement guarantee. Vestis Atlas runs selected shopper questions through connected AI answer engines and records what they return at that time.
How the process works
- Set your brand profile first. In Settings, fill in Brand name, Aliases and Domains. Without a brand name, Atlas can only match your product names and brand mentions are not detected at all.
- Use questions that a real shopper might ask, without forcing your brand name into every question.
- Keep the same prompt set when comparing results. Changing the questions changes the test.
- Look for patterns across several runs. A single AI answer can change from one run to the next.
Go to atlas.vestislabs.com/dashboard, sign in and select "GEO visibility" from the top menu.
Understand what Atlas is measuring
Atlas sends each saved prompt to the connected AI answer engines. It then checks the returned answers to see whether your brand or product is mentioned and stores the result as a dated run.
It gives you a consistent visibility signal: which questions lead to mentions, your overall mention rate and whether that pattern changes over time.
It does not measure every shopper session, website traffic, clicks, sales or a permanent search position. A mention also does not automatically mean that your product was ranked first or positively recommended.
Open GEO visibility and create a prompt set
Enter a clear name for the prompt set, such as "Core products - weekly".
Choose Weekly or Daily. This records your intended rhythm and groups the set; it does not schedule anything. Every run is started manually with "Run now".
Select "Create set".
A prompt set is a saved group of shopper questions that Atlas runs together. Keeping the group consistent makes later comparisons more meaningful.
Choose prompts that give useful results
A prompt is the question Atlas sends to an AI answer engine. The quality of the questions determines how useful the tracking will be.
Use natural wording, product details and real use cases. Avoid including your brand name in every prompt, because that tests whether an engine can retrieve a known brand rather than whether it discovers you.
- describes the product, category, occasion or need clearly;
- is specific enough to produce a meaningful answer;
- matches the products and markets you want to track; and
- can be repeated without rewriting it each time.
Do not substantially edit the prompt set while assessing a before-and-after change. Create a new set when you want to test a different group of questions.
Auto-generate, review and edit the prompts
Select "Auto-generate" to let Atlas suggest product and category questions from your approved products. Products that are still awaiting approval are ignored, so approve the catalogue you want to track first.
- Review every suggested question. Items marked "AUTO" were created by Atlas.
- Correct awkward wording, inaccurate product details or questions that do not suit your market.
- Use the entry field and "Add" to include your own shopper questions. Questions you add yourself are labelled "Category" regardless of wording.
- Use the x at the end of a row to remove a question that is duplicated or not useful.
They make setup faster, but they still require human review. The best set combines Atlas suggestions with questions based on your customers, products and commercial priorities.
Run the visibility check
When the prompt set is ready, select "Run now". Atlas queues the set, sends the prompts to the connected engines and adds the completed results below the prompt list.
- Check that the prompt list is correct before starting.
- Select "Run now" once.
- Wait for the queued run to complete. Results appear when processing has finished.
- Use the back arrow to return to the main GEO visibility overview.
AI answers can change as models, sources and search indexes change. Repeat the same set yourself at the chosen cadence - runs are always started manually - and use the trend across runs rather than treating one result as final.
Selecting "Run now" again queues a second, separate run; nothing blocks it. Each run counts against your plan's allowance for tracking runs and for prompt-by-engine results, so wait for a queued run to finish before starting another.
Read the visibility dashboard
The dashboard summarises the answers Atlas has checked and the mentions it detected across completed runs.
The mention rate shows how often you appeared in the analysed answers. It does not by itself show your position, the tone of the answer, click-throughs or sales.
Compare visibility around enrichment events
Atlas can compare visibility before and after product information is enriched. This helps you see whether stronger, more complete product data is followed by a change in mentions.
- A sustained increase or decrease across several comparable runs.
- Specific prompts that begin or stop producing mentions.
- Product or category areas that remain consistently invisible.
- Large one-off fluctuations that do not continue in later runs.
An enrichment event is a useful reference point, but other factors may also change the answers - including engine updates, newly indexed sources, competitor activity and changing web content.
For a meaningful comparison, keep the prompt wording, prompt set and cadence consistent. Also allow time for updated website data to be published and discovered.
Use the results correctly
AI visibility tracking is most useful as an ongoing directional measure. Use it to identify patterns, prioritise data improvements and test again - not to promise a guaranteed placement in an AI answer.
- Keep a stable core prompt set and run it consistently.
- Record meaningful product-data and website changes.
- Review individual prompts as well as the overall mention rate.
- Base decisions on repeated patterns, not one isolated result.
- Treat the results as evidence to investigate, not proof of cause or guaranteed visibility.
Vestis Atlas measures answers returned through connected answer-engine APIs, which act as a proxy for what shoppers may see. All engines are reached through OpenRouter. Perplexity Sonar runs its own native search stack and is used as the anchor measure. GPT-4o and Gemini 2.5 Flash are run with OpenRouter's web-search plugin, so those results are not necessarily identical to each engine's consumer app. The engine roster can change as models are updated.