How AI Recommendation Monitoring Works

Search engines return links. AI assistants return recommendations.

That sounds like a small difference. It isn't. A buyer types "best project management software for remote engineering teams" into Perplexity. They don't get ten blue links to click through. They get a shortlist: usually three to five brands, named directly, often with a one-line reason for each. The buyer's next step is not research. It's choosing.

AI recommendation monitoring measures whether your brand is on that shortlist.

That is a different question from "how do you rank for this keyword," and it needs a different way of measuring.

What we measure: recommendation presence (how often AI names you)

The core measurement is simple to state and hard to do well. We ask AI assistants the questions buyers ask, and record which brands each answer names.

We record three things: who was named, who wasn't, and who was named instead. Together, they give what we call recommendation presence: how often AI names you when buyers ask for help in your category.

It is easy to confuse with a few other things:

Not brand sentiment. We are not measuring what AI says about your brand when asked about it directly. We are measuring whether your brand comes up unprompted when a buyer asks a category question.

Not search ranking. Search ranking measures where your link sits for a keyword. Recommendation presence measures whether you are on the shortlist for a buyer's question. In practice, the two often don't line up. Brands with strong search authority are often missing from AI shortlists. Brands with modest search presence can show up again and again.

Not mention tracking. Some tools check whether your brand name appears anywhere in AI text. That is a different measure, and a less useful one. A brand mentioned in passing, named as one to avoid, or losing in a comparison is not the same as a brand on the shortlist when a buyer asks who to buy from.

Writing the buyer questions

The questions matter as much as the measurement.

AI assistants give different answers to differently framed questions. "Tell me about Salesforce" is a brand awareness question. "What CRM should a 50-person B2B SaaS company use to manage enterprise pipeline?" is a buyer question. Only the second shows what happens when a buyer is deciding.

Buyer questions are built around:

Category and use case, not brand name. The question never names the brand being measured. We ask who the AI recommends for a use case, not what it knows about a company.

Buying stage. Questions cover the whole buying journey. Early research ("what should I be using to manage X?"), comparison ("how do the leading options for X compare?") and decision ("which option is best for a company in situation Y?"). A brand may be named early but disappear at comparison, or the other way round.

How specific the need is. Vague category questions and specific use-case questions return different shortlists. A brand named for "best email marketing platform" may not be named for "best email marketing platform for a DTC brand with a list under 50,000 that needs deep Shopify integration." That difference tells you something.

A good set of buyer questions for a brand covers several buying stages and levels of detail, scaled to how complex the category is. The overall result across those questions is steadier than any single question.

How we measure

Each question goes to each AI assistant separately. We read every answer for brand names. We record whether the brand was named and, if not, which brands were named instead.

The same questions on every AI assistant. We ask ChatGPT, Claude and Perplexity (plus Gemini on Monitor) the same buyer questions at the same time. Each AI assistant draws on different sources and different snapshots of the web. The same brand can be named often on one and never on another. That gap is a key finding, not noise.

Several runs per question. AI answers vary. The same question asked twice can get different answers. We run each question more than once and report the overall pattern rather than a single snapshot.

Careful name matching. Spotting a brand takes more than a simple text search. We account for brand nicknames, common abbreviations, website names and partial matches. We also note what kind of mention it is: on the shortlist, in a comparison, or held up as an example to avoid.

A plain count. The headline result is a count: the number of answers that named you, out of all the answers across every AI assistant. If a brand was named in 32 of 60 answers, we report 32 of 60. The count is more honest. It shows how many answers sit behind the result, so a handful of answers can't look more certain than they are.

Who is named instead

When your brand is not named in an AI answer, someone else is.

We call that brand the replacement: the competitor in the spot you're not in. For every answer that leaves you out, we record who was named, which competitors do it most often and how often each one does.

Most AI tracking tools don't produce this.

Knowing you were named in 28 of 60 answers tells you the size of the gap. Knowing HubSpot was named in your place in 24 of the 32 answers that left you out tells you which competitor matters most. Those are different strategic questions. Knowing who is named instead turns a single number into a picture of your competition.

The pattern also shows how concentrated a category is. In some categories, one brand is named in nearly every answer that leaves you out. In others, it is spread across five or six competitors. If it's concentrated, there is one main competitor to understand. If it's spread out, the category is fragmented and no single brand leads.

Each AI assistant answers differently

The major AI assistants draw on different sources, update on different schedules and find information in different ways.

Perplexity searches the live web when it answers. Claude answers mainly from what it learned in training, up to a cut-off date. ChatGPT works both ways: with web search on, it searches the live web; with it off, it answers from training. Envoyra asks ChatGPT with web search on, the way most buyers now use it.

These differences show up in the answers. A brand with lots of recent press may be named often on Perplexity and rarely on Claude. A brand with long-standing coverage, such as a detailed Wikipedia page, may do well on Claude and less steadily on Perplexity. We describe these as patterns we observe, not as proof of how each assistant works.

Don't average these differences away. They are part of the finding. A brand named often on every AI assistant has a presence that holds up across sources and over time. A brand named often on one and rarely on the others may be on the rise, or may be on shaky ground.

We report results for each AI assistant separately and flag the differences. Being named consistently across AI assistants is one of the strongest signs of a lasting presence.

Tracking over time

A single scan is a snapshot. AI recommendations change over time.

AI assistants update. New content appears and gets picked up. The sources behind competitors' mentions shift. A brand's presence in March may look quite different in June, even if the brand changed nothing, because the sources around its category changed.

Ongoing monitoring tracks three things:

Movement. Whether a brand is named more often, less often or about the same, and how fast that is changing.

Volatility. How much a brand's presence swings from week to week. Big swings suggest a presence that rests on a few sources that come and go. Small swings suggest a deeper base.

Competitor movement. Whether the brands named in your place are gaining or losing ground. A competitor named in your place more and more often means your position is getting worse, even if your own count holds steady.

Weekly monitoring is built to catch movement early, before it becomes a gap.

What the result means, and what it doesn't

The result of AI recommendation monitoring is a count: your brand was named in X of Y answers across Z AI assistants.

That count is not a grade. It is not a target. It is an observation.

Picture a brand in a narrow, specialist category named in 10 of 15 answers on one AI assistant. Its position may be stronger than a brand in a broad consumer category named in 40 of 60 answers across four. The count only makes sense alongside the category, who is named instead and how things move over time.

What the pattern means for this brand, in this category, right now, can't be boiled down to a number. That's why every Envoyra report explains the result as well as showing it: what it means and where the evidence is most and least firm.

We also state known limits plainly. AI assistants update without notice. Results depend on the questions asked; a different set may give different results. What we observe does not prove cause. We record what AI assistants say; we do not claim to know why they say it. Every result shows how many answers it is based on.

Why this matters now

The brands showing up in AI answers today may be building a lead. Brands that are named consistently tend to keep being named, and brands that are missing tend to stay missing while competitors settle in. We report this as a pattern we observe, not a rule.

Most brands don't know where they stand. They assume that doing well in search means doing well in AI. That assumption is wrong often enough to be dangerous.

The first step is measurement. Not strategy. Not rewriting content. Measurement: an honest look at where you stand in the answers that now come before every other channel you've worked on for years.

Request an AI Presence Audit

ChatGPT, Claude and Perplexity, plus Gemini on Monitor. Real buyer questions. See who is named instead of you.