How we measure AI visibility

1. Tracked prompts, not keywords

Classic SEO tracks a keyword to a rank position. AI answer engines return one generated paragraph, so there is no position to scrape. DeepLexa reframes the problem: we ask the AI the buyer-intent questions your customers actually ask, and measure whether you appear. Your "keywords" become tracked prompts — questions like "best project management tool for remote teams", "alternatives to Notion", or "is Acme any good".

2. Multi-engine querying

We run each prompt across multiple engines in parallel. Perplexity (Sonar) is our highest-fidelity engine because it queries the live web and returns citations, and its API is the same engine as its consumer product. OpenAI captures ChatGPT's parametric knowledge, and Google Gemini adds another major model. Google AI Overviews and further engines expand coverage over time.

3. Presence detection

For every answer we detect four things: mention (does your brand appear?), citation (does your domain appear as a source?), prominence (how early in the answer are you named?), and the competitors named instead of you. This competitor signal is often the most actionable part of the report.

4. The Share of Answer score

We aggregate into a single 0–100 score, weighted approximately 0.65 × mention rate + 0.20 × citation rate + 0.15 × prominence when citation data is available. You also get a per-engine breakdown, the specific prompts where you're invisible ("gaps"), and your top competitors. Failed engine calls are never rendered as a fake "0" — a failed scan shows an honest error and fabricates no score.

5. Prioritized fixes

Gaps and competitor data are converted into a ranked action list across five categories: content, entity, citations, comparison pages, and technical/retrievability. Each fix is written in plain English and ordered by the visibility it is likely to unlock.

6. Tracking over time

A weekly scan re-runs the same prompt set and stores each result, producing a trend line over the same questions and alerting you when your visibility moves or a competitor overtakes you.

Honest limitations

• Non-determinism: identical prompts can return different brands, so we track rates over time, not single lookups.
• API vs consumer UI: API answers approximate but do not perfectly match the logged-in consumer experience; Perplexity is closest.
• Sampling: we sample a prompt set — the value is trend and competitor signal, not one absolute number.

We believe being honest about these limits is what makes the numbers trustworthy. Measuring AI visibility imperfectly and acting on the trend beats not measuring it at all.