AI Prompt Observatory

See whether AI answer engines recommend your brand

The Prompt Observatory asks your buyers’ questions to the AI engines every week and records which brands get named. When your customers ask an AI assistant a buying-intent question, you find out whether the answer includes you — or only your competitors.

Today every question goes to two engines: Perplexity, which answers from the live web and cites its sources, and Claude, which answers from model knowledge alone. That pairing is deliberate — it separates what an engine can look up about you right now from what it already believes about your category, and those two answers are often not the same.

  • Your share of model over time — tracked weekly, not guessed once from a single spot-check.
  • Who got named instead of you — by name, for each question.
  • Which sources the answer cited — the pages an engine actually read before it recommended someone else.
  • Alerts only when the answer changes — a weekly email that always fires gets muted, so ours doesn’t.

Tracking is included in Pro at $49 a month. Not ready yet? Run the free AI visibility check first — it takes seconds, needs no signup, and tells you whether engines can read and quote your pages at all. Then come back and watch it.

New to this? AI brand monitoring explains what to track and how often, and answer engine optimization covers how to fix what the tracking finds.

Asked again every week

Put your questions on the schedule

Pick the buying-intent questions that matter, name the competitors you care about, and let the Observatory re-ask them every week. You get the movement, not another dashboard to remember to open.

Start tracking — $49/mo →How Share of Model works


Preview of the AI Prompt Observatory with Website URL and options fields and a Result result

Why two engines rather than one

Perplexity and Claude fail differently, and that is the point of running both. Perplexity answers from the live web and cites what it used, so it reflects what is findable and well-sourced right now. Claude answers from model knowledge alone, so it reflects what has settled into the training data over a longer period. A brand can be named by one and absent from the other, and which one tells you what kind of problem you have.

Being named by Perplexity but not Claude usually means the material exists and is reachable but has not yet accumulated enough independent coverage to be remembered. The reverse, named by Claude but missed by Perplexity, points at a retrieval problem: the reputation exists but the current pages are not being found or cited. Tracking only one engine collapses those two very different diagnoses into a single number.

Illustration: prompts feeding a dashboard that tracks AI mentions and share of voice over time

Choosing questions worth tracking

The questions you schedule determine whether the results are actionable or merely interesting:

  • Use the questions buyers actually type, not the ones that flatter you. Best tool for X is worth tracking; is DigiJaws good is not, because nobody asks it.
  • Avoid naming your own brand in the question. A prompt that contains your name guarantees a mention and measures nothing.
  • Include the comparison questions. X versus Y and alternatives to X are where purchase decisions are actually made.
  • Cover the problem as well as the category. Many buyers describe a symptom before they know the category name, and those answers name a different set of brands.
  • Keep the wording fixed once tracking starts. Rewording a prompt resets the trend, because you are no longer measuring the same thing week to week.
  • Track enough questions to see a pattern but few enough to read. A dozen well-chosen prompts beats fifty you never look at.

Frequently asked questions

How often do the questions run?

Weekly. That is frequent enough to catch a genuine shift in how engines describe your category, and infrequent enough that you are not reacting to the normal variation between individual answers. Because the same prompts run on the same schedule, the value builds over time: a single run is a snapshot, while a quarter of runs shows direction.

Which AI engines are queried?

Perplexity and Claude. They are chosen because they answer in fundamentally different ways. Perplexity retrieves from the live web and cites its sources, so it reflects what is currently findable. Claude answers from model knowledge without browsing, so it reflects what has settled into the training data. Together they separate a retrieval problem from a reputation problem.

Why do the answers change between runs?

Because these systems are not deterministic. The same prompt can produce different wording, a different ordering of brands, and sometimes a different set of brands entirely, even with nothing changed on your side. That is why a single run should never be treated as a verdict. The signal is the rate at which you are named across repeated runs, which is exactly what tracking the same prompts on a schedule measures.

How many questions should I track?

Ten to fifteen is a good working range for most categories. That is enough to cover the main buying questions, the comparisons and a couple of problem-led phrasings, while staying small enough that you will actually read the results each week. Adding more prompts mostly adds noise, unless they open a genuinely different part of the market.

Free tools →My toolkit →Locations →Comparisons →
Verified Agent-Ready by DigiJaws
See who AI recommends instead of you — free AEO audit, 90-day plan you keepGet my free audit