TL;DR
Key Definition
An AI mention occurs when an AI assistant names your brand in its answer, for instance inside a list of recommended tools. An AI citation goes further and points to a source page with a link. Mentions draw on two pools, what the model learned in training and the pages it pulls live from the web. How often you appear, where and in what tone tells you how AI assistants pitch your brand to your prospects.
Importance
Why Does It Matter?
- A buyer asks an AI which tools to pick. Only the brands it names make the starting shortlist. For that buyer, the rest do not exist.
- A mention often comes with no link. It stays invisible in your analytics, and only direct measurement of the answers brings it to light.
- Context is everything. Being framed as the right choice for a specific need counts for far more than a spot at the bottom of a list.
- A wrong or dated description of your offer, say an old price or a retired feature, gets repeated to every user who asks the question.
- Comparing your mentions with your competitors on the same questions shows where you lose prospects before they even open a website.
How Does It Work?
A language model names a brand when it looks relevant to the question. To decide, it draws on what it learned in training. If web search is on, it also draws on the pages it retrieves at answer time.
- 1
Model Memory and Live Search
Researchers separate parametric memory, held in the model weights, from non-parametric memory, made of documents retrieved on demand (Lewis et al., 2020). A brand named without web search comes from the first. It reflects how present the brand was in training data, and that picture may be dated. With web search, the model reads recent pages, so mentions respond more to your current content and to third-party articles.
- 2
Types of Mentions
A mention takes several forms. A recommendation ("for a 200-person sales team, tool X is the best fit"), a single line in a list, a comparison against a competitor, an answer to a direct question about the brand. Its tone can be positive, neutral or negative. Sitting in a list of ten tools carries less weight than being recommended in the very first sentence.
- 3
Where the Model Gets Its Information
Your site is rarely the only source. Comparison articles, reviews, forums, documentation, trade press and Wikipedia pages describe your brand in their own words. When those sources agree, the model repeats a stable description. When they clash or have aged, the mention turns vague or flat out wrong.
- 4
Why Mentions Vary
Ask the same question twice and you may get two different lists. Generation involves some randomness, and search results move. Platform, language and wording matter too. That's why you measure a frequency over a sample of repeated prompts, never a single screenshot.
Metrics
How Do You Measure It?
- Build a prompt set that mirrors your buyers, from discovery questions through to comparisons, and run it at regular intervals on ChatGPT, Gemini, Claude and Perplexity.
- Compute the mention rate platform by platform, dividing the number of answers that name your brand by the number of answers analyzed.
- Also log the brand rank in each answer and the tone used (recommended, neutral, criticized). Quality matters as much as volume.
- Compare your mention rate with your competitors on the same prompts to get your share of voice.
- List the factual claims made about your brand (price, features, customers) and check them against reality to catch errors.
- Hikoo's Agent reporting tracks these mentions, their sentiment and share of voice against competitors on your prompt set.
Pitfalls
What Are the Common Mistakes?
- Drawing conclusions from a single test on a handful of questions, when answers shift from one run to the next.
- Counting mentions without reading what surrounds them. A recommendation and a criticism end up on equal footing.
- Tracking only prompts that already include the brand name. Your prospects mostly ask questions with no brand in them.
- Overlooking the third-party sources that talk about you, when the model leans on them heavily.
- Letting contradictory information float around your site, directory listings and profiles, which muddies the description AI assistants repeat.
- Leaving competitors out. You then can't tell whether a drop in mentions comes from you or from a new entrant.
What Should You Do First?
Follow these steps to get started.
- List the variants of your brand and product names to look for in answers.
- Write 30 to 50 unbranded prompts modeled on your buyers' questions.
- Run that set every month on at least three AI assistants and archive the answers.
- For each answer, note whether the brand appears, at what rank, in what tone and which competitors are named.
- Fix errors at the source, on your site and on third-party pages you can get updated.
- Make the description of your offer match across your site, profiles and partner listings.
- Publish clear pages on the cases where your product is the right choice. The model will then have grounds to recommend you.
Examples
Recommendation in a List
Illustrative example. Someone asks ChatGPT "which CRM for a network of 40 real estate agencies". It names five tools and recommends one for its listing management. The other four get a mention, but only one is singled out, and that one stands the best chance of getting the visit.
Outdated Price Repeated by AI
Typical scenario. A SaaS vendor notices that several assistants quote an entry price it dropped a year ago. The figure comes from an old comparison article that got reused everywhere. The team updates its pricing page, writes to the article author and watches the answers over the next few months.
Mention Rate by Platform
Hypothetical example. Across 40 tracked prompts, a brand is named in 22 Perplexity answers against only 9 on ChatGPT. Given the gap, it checks OAI-SearchBot access to its site and builds up its presence in the comparison articles ChatGPT cites most.
Frequently Asked Questions
What is the difference between an AI mention and an AI citation?
An AI mention names your brand in the text of the answer. An AI citation points to a page as a source, usually with a link. The first builds awareness and gets you onto the buyer's shortlist. The second adds a potential click to the cited page, which isn't necessarily yours.
How can I increase my AI mentions?
Make your brand easy to describe and easy to recommend. Publish pages that spell out who your product is for and which need it solves best, and keep your information consistent everywhere. Get featured in the third-party sources AI assistants read, whether comparison articles, trade press or reviews. And let AI search crawlers into your site.
How do I know if ChatGPT mentions my brand?
Ask ChatGPT your prospects' questions without naming your brand, and repeat each one several times since answers vary. Note whether the brand appears, at what rank and in what terms. To follow several platforms over time, a prompt tracking tool spares you hours of manual testing.
What should I do about a negative or wrong AI mention?
Start by finding the likely source, in the cited pages or by searching for the claim on the web. Then fix the information on your site, ask for wrong third-party pages to be updated and respond to fair criticism. Models with web search pick up corrections faster than their training memory does.
Are AI mentions without a link worth anything?
Yes, because they put your brand on the buyer's shortlist right when they compare. Many prospects then type your name into Google or enter the URL directly. That traffic shows up as branded or direct rather than as an AI referrer. So keep an eye on your branded searches alongside.
Sources
- 1arXiv GEO, Generative Engine Optimization (Aggarwal et al., 2023)
- 2arXiv Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (Lewis et al., 2020)
- 3OpenAI Platform Docs Overview of OpenAI Crawlers
- 4Anthropic Help Center Does Anthropic crawl data from the web, and how can site owners block the crawler?
