A SaaS company runs a batch of prompts through ChatGPT. Its name shows up in 41 percent of the answers about “best project management software.” Then it asks a sharper question: “which one should I buy.” A different brand wins every time. Never the company running the prompts. That gap, mention share against recommendation share, is the number most AEO reports leave out of the deck.
Key Takeaways
- Mention share measures how often a brand appears in AI answers. Recommendation share measures how often it is named as the actual pick, and the two move independently.
- Models retrieve a wide set of candidates before they reason about which one best fits the query, so a brand can clear the retrieval bar constantly without ever clearing the recommendation bar.
- Because citations compound, a brand stuck as the “also consider” option tends to stay there, since each prior answer reinforces the next one.
Showing Up Isn’t the Same as Winning
Share of Citation tells you how often a model pulls a brand in and names it as a source. Share of Voice is blunter: it just counts mentions anywhere on the web. Recommendation share sits past both of those. It asks one question only: did the model frame this brand as the answer, or did it just leave it sitting on a list?
Three different events, one response. A model can retrieve a brand’s page, cite it as a source, list it beside five competitors, and still close the answer with someone else’s name as the pick. Count the citation and the brand looks strong. Count the recommendation and it’s nowhere.
Why a Pile of Citations Doesn’t Buy the Top Spot
A model answering a comparison query retrieves dozens of documents that mention the category, pulls candidate brands out of them, then checks those candidates against a trust graph: do independent, authoritative sources actually agree on what fits the question being asked? Clearing the retrieval step is easy. Directory listings, roundup mentions, thin blog posts, any of it lands a brand in the pile.
Getting named as the recommendation asks for something harder. Convergence, not acknowledgment. Twenty roundup articles that each list a brand as one of ten options add up to very little. Three independent, high-authority sources that each single it out for a specific use case add up to a lot, even with a fraction of the raw mention count behind them.
What NewDay’s Rank Data Actually Shows
This is the line between a visibility number and a commercial one. Raw mentions say a brand is in the conversation. Rank and recommendation data say whether the model is actually sending people toward it.
average rank on Google AI Overviews achieved by NewDay USA, alongside 64 percent growth in overall AI brand presence, tracked weekly across five engines over six months.See the case study
An average rank near 1.6 means something specific. It means the model consistently put the brand first or second when the question actually mattered. That’s the number that tracks with revenue. Mention volume alone wouldn’t have caught it.
The Compounding Trap: Stuck as Runner-Up
Citations compound. Once a model cites a brand, the next citation gets more likely, because the model leans on patterns of agreement it has already seen rewarded. The same mechanism runs in reverse for a brand that keeps getting cited but never picked.
If a model has learned to list a brand third in “best X” answers, that pattern holds. The training data and retrieval results behind it don’t rewrite themselves. So the mention count climbs quarter over quarter while the actual recommendation never moves an inch. A team watching only total citations reads that as growth. It’s stagnation with a growth chart taped over it.
A brand can be named in every answer and chosen in none.
What to Track Instead of Raw Mentions
Splitting these two numbers means splitting the prompt set by intent, not by topic. Informational queries (what is X, how does X work) surface mentions regardless of commercial fit. Decision queries (which should I use, what do you recommend for Y) are where recommendation share actually shows up, and where it needs its own tracking line.
Each engine reaches that decision differently, so the split runs per engine, not as one blended number. ChatGPT draws from across the open web, which rewards broad coverage and dilutes any single source’s pull. Gemini cross-references Search, YouTube and Scholar, so video and academic presence carry more weight than they would elsewhere. Claude leans on high-authority publications and documentation, rewarding fewer, stronger citations over a pile of weak ones. Perplexity cites its sources inline, which makes it the easiest engine to audit directly for whether a citation turned into a recommendation. Grok reads the real-time social web, where recency and sentiment move the outcome week to week.
In practice: log every answer where the brand gets mentioned, tag whether it was framed as the pick or as an alternative, and track that ratio by engine and by query intent over a rolling period. A rising mention count next to a flat or falling recommendation ratio is the signal worth acting on. The mention count by itself isn’t.
FAQs
What Is the Difference Between Mention Share and Recommendation Share?
Mention share counts how often a brand turns up anywhere in an AI answer, including as one option among several. Recommendation share counts how often the model names that brand as the actual pick for a decision-oriented query. A brand can score high on one and near zero on the other.
Why Would a Model Cite a Brand Without Recommending It?
Models retrieve a wide set of candidates first, then weigh them against a trust graph of independent, agreeing sources. A brand with broad, shallow coverage clears retrieval easily but may never build enough converging authority to get singled out as the recommendation.
How Can a Brand Track Recommendation Share Separately From Mention Share?
Split the prompt set by intent: informational queries versus decision queries that ask which option to pick. Tag each appearance as either the pick or an alternative, and track that ratio by engine over time, since each engine weighs sources differently.
Does a Higher Mention Count Always Mean More Revenue?
No. Mention count can climb steadily while recommendation position stays flat or slips, because citations compound in whatever pattern the model already learned. Revenue tracks closer to recommendation share and rank position than to total mentions.
