Recommended vs Mentioned in AI Answers: Why It Matters
Being named in an AI answer is not the same as being picked. How recommended, mentioned, negative and cited differ, and why a mention rate hides it.
- ai visibility
- brand mentions
- citations
- geo
Recommended vs mentioned is the difference between an AI answer that tells a buyer to pick you and one that only lists you among others. SparkToro found City of Hope named in 97% of ChatGPT answers but the top mention in only 25 of 71, so a mention rate alone cannot tell you which of the two your buyers hear.
What does the difference look like inside one answer?
An AI answer can treat your brand in four distinct ways, and only one of them is the outcome you want. Take a buyer asking an assistant for a CRM for a 10-person team. These four sentences could all appear in answers to that same question. (Invented example with fictional brands.)
| Outcome | Example sentence | What it does for you |
|---|---|---|
| Recommended | "For a 10-person team I would go with Northwind CRM: simple pipeline, EU hosting." | Puts you on top of the shortlist |
| Mentioned | "Options include Northwind CRM, Contoso and Globex CRM." | Keeps you in the set, decides nothing |
| Negative | "Northwind CRM is cheap, but users report slow support, so be careful." | Names you and pushes the buyer away |
| Cited | A source link to northwind.example, whether or not the text names the brand | Shows the assistant read your page |
A recommendation is an answer that actively puts your brand forward as the choice or the top option. A mention is your name in the answer text, neutrally, as one of several options. A negative mention warns against you. A citation is a link to your own domain as a source; Nodpick's brand mention glossary entry notes that "You can have one without the other".
Counting all four as "visible" is how a brand can look healthy in a dashboard and still lose deals. Here is how much that can hide.
How much can a mention rate hide? A ten-answer test
Two brands with the same mention rate can have nothing else in common. In the worked example below, both brands are named in 6 of 10 answers, a 60 % mention rate. One is recommended in 4 of them; the other in none, and warned against once.
| Ten answers to the same buyer questions | Brand A | Brand B |
|---|---|---|
| Named in | 6 of 10 (60 %) | 6 of 10 (60 %) |
| Recommended in | 4 (40 %) | 0 (0 %) |
| Negative | 0 | 1 |
| Own domain cited | 2 | 0 |
| Position score (100 / 80 / 60 / 40 / 0) | 52 | 40 |
Method. Ten invented answers per brand, each labelled with the brand's position, how the answer treats it and whether its domain is cited; position points as Nodpick publishes on its method page (first named 100, second 80, third 60, later 40, not named 0). Brand B's one first place sits in a neutral list, not a "best of" list. Computed 31 October 2026. Limits. A constructed example to show the arithmetic, not measured data.
A mention report shows these brands as equals. A position score separates them a little, 52 against 40. Only the recommendation count shows that one brand wins buyers and the other sits in lists.
Ten answers are also too few to be sure of the gap. Brand A's 4 of 10 recommendations carry a 95 % Wilson range of 17 to 69 %; Brand B's 0 of 10 still allows up to 28 %. The same 40 % rate from 100 answers narrows to 31 to 50 %. A recommendation rate needs the same breadth of questions as a mention rate, and real answers add another problem: position itself is unstable.
Is position a good stand-in for being recommended?
Position helps on average but is too unstable to stand in for a recommendation in a single answer. SparkToro found two lists in the same order "more like 1 in 1,000 runs", and Rand Fishkin's verdict on the practice is blunt: "any tool that gives a 'ranking position in AI' is full of baloney." (SparkToro ran that study with Gumshoe.ai, a startup that sells AI tracking, and names the possible conflict of interest in the post.)
Averaged over many answers, being named earlier still carries a signal, which is why Nodpick's visibility score weights position across all answers and shows a range, instead of reporting a rank. But an answer can name you second and still recommend you ("Contoso if you need invoicing, otherwise Northwind CRM is my pick"), or name you first and then warn against you. Whether you were recommended is in the wording, not the order.
So the wording matters. Does it matter if nobody clicks anyway?
Why does the wording matter if buyers never click?
Because for most buyers the answer is the decision, not a step before a visit. Pew Research tracked the browsing of 900 U.S. adults in March 2025. Users who saw a Google AI summary clicked a traditional result in 8% of visits, against 15% without a summary, and clicked a link inside the summary in just 1% of visits.
Pew also found that users ended their browsing session on 26% of pages with an AI summary, compared with 16% of pages with only traditional results. A buyer who reads "go with Northwind CRM" and closes the tab has made a choice your analytics will never record. A recommendation reaches that buyer; a neutral mention barely does; a negative mention reaches them with the wrong message.
Which raises the question many teams start from: are citations not the real prize?
Is being cited worth as much as being recommended?
A citation shows the assistant read your page; a recommendation shows what it concluded. Both matter, but they do different jobs, and one does not imply the other. Google says AI Overviews and AI Mode "may use a 'query fan-out' technique - issuing multiple related searches across subtopics and data sources", so a single answer can draw on many pages, yours among them, and still recommend someone else.
| Cited, not recommended | Recommended, not cited | |
|---|---|---|
| What happened | Your page was a source, the conclusion went elsewhere | Third-party pages made the case for you |
| What to check | Does your page state clearly who you are best for? | Which sources does the assistant cite for you? |
| Typical next step | Sharpen positioning and comparison content | Keep and widen presence on those third-party pages |
The best case is both: your own page cited and your brand recommended. The worst case is a citation of your page in an answer that recommends a competitor. Telling these apart at scale needs a consistent rule for what counts as a recommendation.
How does a tracker decide which one it was?
A tracker decides "recommended" from the wording of the answer, so its rules decide your number. The rules below work as a checklist for any tool or spreadsheet; Nodpick's method page explains its three groups with examples.
| Rule | Counts as recommended | Does not count |
|---|---|---|
| Recommending words close to your name | "I'd recommend Acme", "Acme is the best choice for small teams" | Recommending words that sit closer to another brand |
| Top of a "best" list | First place in a list framed as the best options | First place in a neutral list of options |
| Heading or conclusion | A conclusion that names you as the pick | A plain description of your company |
| Someone else's pick | "we recommend HubSpot over Acme" counts for HubSpot, not for Acme | |
| Warning words | "avoid", "not recommend": sorted as negative |
Nodpick reads every answer that names you and sorts it into recommended, just mentioned or negative, counting only brands that actually appear in the answer. A reading of the text is not a reading of intent, so every answer is listed with its group and you can check each call yourself.
How do you move from mentioned to recommended?
Start with the answers where you are only mentioned, and read who is recommended instead and which sources the assistant cites for them. Those sources are often lists, reviews and comparison pages. Content changes can move the result: the GEO study by Aggarwal et al. (first version 16 November 2023) reports visibility gains in generative engine responses "by up to 40%".
Work in this order:
- Find the gap. Questions where you are named but another brand is recommended.
- Read the sources. The domains cited next to the recommended brand show where the case is made.
- Fix the strongest-evidence item first. Being reachable by AI crawlers is documented by the platforms; much other GEO advice is untested.
- Re-measure the same questions and wait for a change that beats normal week-to-week noise.
Doing this by hand means reading every answer and labelling it yourself, every week, with rules you keep consistent. Nodpick sorts every answer that names you into recommended, just mentioned or negative, records separately whether your own domain is cited, and shows your recommendation rate with a 95 % range next to the score, or "unclear" with the plain count when fewer than 10 answers back it. It shows who is named instead of you and which sources the answers cite, ranks fixes with an evidence grade (A documented, B studied, C untested), and calls a change real only when it passes a written noise rule. Nodpick opens soon: join the waitlist for early access and the founding price, in one email the day it opens.
Common follow-up questions
Does a negative mention hurt more than no mention?
Often, yes, for that one question: the buyer hears your name together with a reason not to choose you. It is also the most actionable result, because the answer usually says why. Read the cited sources behind the warning; outdated reviews or one loud complaint thread are common causes and can be addressed.
Is "recommended" the same as positive sentiment?
Not quite. Sentiment describes the tone around your name; a recommendation is a decision in the answer. An answer can describe you warmly and still pick a competitor, or list mild drawbacks and still recommend you. Track the decision first and the tone second.
Can I be recommended without being cited?
Yes. An assistant can recommend a brand on the strength of third-party lists, reviews and directories without linking the brand's own site. That is still a strong outcome. It also tells you which outside pages carry your reputation, so keeping them accurate matters as much as your own pages.
How do I measure a recommendation rate by hand?
Ask a fixed set of unbranded buyer questions, read every answer, and label each one recommended, mentioned, negative or not named, plus whether your domain is cited. Divide recommended answers by all answers. Keep wording and country fixed, and repeat weekly, because single answers vary from run to run.
Sources
- SparkToro, "NEW Research: AIs are highly inconsistent when recommending brands or products" (27 January 2026), https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/, Last checked: 31 October 2026
- Pew Research Center, "Google users are less likely to click on links when an AI summary appears in the results" (22 July 2025), https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/, Last checked: 31 October 2026
- Google Search Central, AI features and your website, https://developers.google.com/search/docs/appearance/ai-features, Last checked: 31 October 2026
- Aggarwal et al., "GEO: Generative Engine Optimization", arXiv:2311.09735, https://arxiv.org/abs/2311.09735, Last checked: 31 October 2026
- Nodpick, How we measure AI visibility, https://nodpick.com/method, Last checked: 31 October 2026
- Nodpick, Glossary: Brand mention, https://nodpick.com/glossary/brand-mention, Last checked: 31 October 2026
- Worked example (ten labelled example answers per brand), computed 31 October 2026
Checked 31 October 2026.