How Envoyra measures AI recommendation presence

The three steps above are the machinery. Around them runs a five-step loop we repeat with every customer, week after week: Focus, Observe, Contextualize, Act, Learn. We call it the FOCAL loop.

You set the focus. Our scans observe: they ask AI assistants the questions your buyers ask and record the answers. Together we work out what those answers mean for you (Contextualize). Acting on them is your decision and your team’s work. The following weeks show what changed, and that learning sharpens the next round.

We split the work this way on purpose. Because acting is yours, our measurement has nothing to defend. We measure the same way whatever you do next, and that keeps the numbers honest.

Envoyra sends each buyer question straight to the AI assistants and records every answer. With each answer we store which AI gave it, the model version, and the time. So any result can be traced back to its source.

If you audit our data, you can check exactly what we asked, when, and under what conditions.

Analytics hold no personal data. Free scan data never enters our research data.

Envoyra measures recommendation presence: out of all AI answers to buying questions, how many name your brand as an option. Not brand mentions in general text. Not search rankings. Not social reach. Only answers to someone who is deciding what to buy.

We ask AI assistants buying questions and record which brands each one names. We record who is named, who isn’t, and who is named in the answers where your brand is missing. We track how this changes over time, and whether it differs from one AI assistant to the next.

We write the questions the way real buyers ask AI: about a category or a need, not about a brand by name. “Best project management tools for remote teams,” not “tell me about Asana.” The first measures discovery: does AI bring you up on its own? The second only measures whether AI knows you.

We record what AI says. We do not try to work out its training data, predict its behavior, or claim to know why a brand is named or not.

AI answers vary. Ask the same question twice on the same day and you can get two different answers. So one answer is one sample, not a fixed result. Turning a handful of answers into a precise-looking percentage would suggest more certainty than the data has.

We report counts: 7 of 25 answers. A count shows how many answers the result is based on. A percentage hides it.

Monitor collects 100 answers per focus area (a topic we track) each week. Once a sample has more than 50 answers, we use a standard statistical method, the Wilson confidence interval, to show a range instead of one number. The range shows how much a result could move by chance.

We label week-to-week changes as up, down, or unchanged. We do not call small changes significant. A change of 3 answers out of 75 is normal week-to-week wobble for most brands. We separate real change from noise, and we do not treat every change as a trend.

We never show percentages anywhere in the product. This is a rule about honest measurement, not a design choice. Counts show what the data actually supports.

Every Envoyra scan reports whether it ran fully, next to its results. We show you this because “AI did not name your brand” and “the scan did not run properly” mean very different things. Mixing them up leads to wrong decisions.

Different AI assistants often give different answers to the same questions. A brand can be named in 18 of 25 Perplexity answers and 4 of 25 Claude answers for the same set of questions. That gap is a real finding, not noise.

Each AI assistant gathers information differently, and the results show it. A brand named often by one and rarely by another is in a different situation from a brand missing from both.

We report each AI assistant’s results separately and call out large gaps between them. Blending them into a single “AI visibility score” would hide what matters. We do not blend them.

With 100 answers per focus area each week (25 questions, each asked of 4 AI assistants), Monitor reports a change only when it is bigger than the range the Wilson method allows for chance. Small week-to-week wobbles usually aren’t, and we don’t call them trends. How big a change must be depends on how many answers we have. That is one more reason we always show the count.

A single scan is a snapshot. How often AI names a brand one week may differ the next week, even if nothing about the brand changed. AI answers simply vary. The picture steadies as more weeks of scans build up. Treat a single scan as a starting point, not a final answer.

We cannot explain why AI recommends what it does. We see patterns that tend to go together, such as which websites are cited and how often a brand is named. We have no access to training data, search logic, or model internals. Any product that claims to explain AI ranking factors is claiming more than the data shows. We don’t make those claims.

We cannot promise that results on our questions hold for every way a buyer might ask. Our questions cover each buying stage, from first question to purchase, but no list is complete. Questions we did not ask may give different results. The right reading is “on these questions, we saw X,” not “everywhere, X.”

Envoyra is a measurement company. We do not sell better numbers.

A company that measures you and also promises improvement wants your numbers to rise, because rising numbers justify the next invoice. Over time, that pull weakens honest measurement. We measure what AI assistants actually say, not what would make the nicest trend line on a dashboard.

Brands that AI names often tend to be covered by trusted independent sources and described the same way across the web. We can show you where AI rarely names you, who it names instead, and which patterns tend to go with being named. What you do with that is your decision and your team’s work. When we advise, the advice cites the same numbers you already have, and the measurement is never tuned to justify it.

We measure the same way whatever you do next. That keeps the numbers honest.