You can't directly make an AI engine mention your business, and the question itself points to a misunderstanding. ChatGPT, Perplexity, Gemini, and Google's AI Overviews don't place businesses into answers the way an ad network places ads or a search engine ranks pages. They name businesses they can describe with confidence and find confirmed by independent sources. The practical goal is not to target the AI. It is to become the business whose description is clear, consistent, and verifiable wherever a machine might look.
Why "Getting Mentioned" Is the Wrong Mental Model
Most advice about AI visibility borrows from search engine optimisation. It assumes there is a ranking to climb, a signal to send, or a file to publish that will tip the system in your favour. That framing doesn't fit how AI answers are produced.
When a user asks an AI assistant for a recommendation, the system writes an answer from patterns in its training data, from pages it retrieves in real time, or from both. Naming a specific business is a small risk for the system. If the name is wrong, outdated, or poorly matched to the question, the answer gets worse. So models tend to name businesses that are:
- Unambiguous: the model can say clearly what the business is and does.
- Corroborated: several independent sources say roughly the same thing.
- Relevant: the business clearly fits the specific question being asked.
Seen this way, an AI mention is less like a placement and more like a summary of agreement across sources. You can't buy that agreement or force it. You can make it easier for that agreement to form.
The Two Ways AI Systems Learn About a Business
AI systems draw on two distinct sources of knowledge. Each responds to different actions on different timelines, and confusing them leads to wasted effort.
| Training Data (Parametric Knowledge) | Live Retrieval (Search-Grounded Answers) | |
|---|---|---|
| What it is | What the model absorbed from its training data before its knowledge cutoff | Pages the system fetches from a search index when answering a question |
| Speed of change | Slow. Updates depend on new model versions, typically months to years apart | Fast. New or updated pages can appear in answers within days or weeks |
| What influences it | A broad, long-standing footprint across the web: coverage, discussion, references | Crawlability, search index presence, and content that is easy to extract and quote |
| Where it shows up | Answers produced without web search | Perplexity, ChatGPT with search, Google AI Overviews, Copilot |
| Main risk | Outdated or missing information about newer businesses | Being blocked from crawlers, or publishing content that can't be cleanly extracted |
For most businesses, live retrieval is the nearer-term opportunity, because it responds to changes you can make now. Training data reflects your accumulated reputation over time. Both reward the same underlying quality: a clear, consistent, well-supported description of what you do.
What an AI System Needs Before It Will Name You
Four conditions determine whether a business is likely to appear in an AI-generated answer. Missing any one of them weakens the others.
- Describability. The model needs a confident, specific sentence about you: what you do, for whom, and where. "A bookkeeping firm for independent restaurants in Melbourne" can be described. "An innovative financial solutions partner" cannot.
- Corroboration. That description should appear across sources you don't control, such as reviews, directories, industry publications, comparison articles, and community forums. Agreement among independent sources is what allows a model to state something as fact rather than hedge.
- Retrievability. For search-grounded answers, your pages must be crawlable and readable. That means important information in plain HTML rather than locked in images, PDFs, or scripts, and no accidental blocking of search or AI crawlers.
- Query fit. The business must clearly match the question asked. Mentions depend on the question. A business can be absent from "best accounting software" and the obvious answer to "accounting software for nonprofit grant tracking."
What Businesses Should Actually Do
The useful work is mostly unglamorous. It focuses on clarity and corroboration rather than tricks.
- Write your one-sentence description and use it everywhere. Decide exactly what you are, for whom, and where. Then align your website, social profiles, directory listings, and business profiles so they all say the same thing. Inconsistency is one of the most common and least noticed reasons businesses go unnamed.
- Map the questions you should legitimately answer. List the specific questions a potential customer might ask an AI assistant where your business would be a genuinely good answer. Narrow questions are where smaller businesses can compete.
- Publish information worth quoting. Pricing ranges, service areas, specifications, honest comparisons, and clear explanations of how you work give retrieval systems concrete material to cite. Put the answer first on the page, not buried under introductions.
- Earn third-party descriptions. Encourage detailed reviews, correct your directory listings, contribute to industry publications, and take part in community discussions where your expertise is relevant. What others write about you is often weighted more heavily than what you write about yourself.
- Check technical access. Confirm that your robots.txt file and hosting settings don't unintentionally block search engine crawlers or the crawlers AI search tools use to fetch pages.
- Be cautious about shortcuts. Files such as llms.txt, schema markup on its own, or pages stuffed with keywords may do no harm, but none of them substitutes for consistent, corroborated information. No major AI provider has confirmed treating llms.txt as a factor in deciding which businesses to mention.
How to Tell Whether It's Working
AI answers are not fixed. The same question can produce different businesses on different runs, on different platforms, and for different users. A single test, where you ask ChatGPT once and see if you appear, tells you very little.
A more reliable approach:
- Build a set of 20 to 50 realistic questions, ranging from broad to highly specific.
- Run them repeatedly across several AI platforms at regular intervals.
- Track how often you appear (your share of mentions), not just whether you appear.
- Note how you are described. An inaccurate description is a signal that conflicting information exists somewhere the model can see.
The way an AI system describes your business is also a useful diagnostic. It shows you what the web, taken together, currently says about you.
What Most People Miss
Most businesses treat AI invisibility as an exposure problem. They assume the model hasn't seen them, so the fix must be more content and more pages. Often the model has encountered them. It just can't describe them with enough confidence to include them in an answer. A business that calls itself an "innovative solutions partner" on its site, a "consultancy" on LinkedIn, and an "agency" in directories sends three weak, conflicting signals instead of one strong one.
The second overlooked point is that self-description carries limited weight. A model producing a recommendation is effectively looking for agreement. A claim that appears only on your own website is a single source. The same claim repeated by a trade publication, a comparison article, customer reviews, and a forum thread becomes something the model can state plainly. Your website matters most as the authoritative version that other sources echo.
The third is that mentions are not a yes-or-no status. No business is simply "mentioned by ChatGPT." It is mentioned for certain questions, sometimes, on certain platforms. Specialized businesses rarely win broad "best X" questions, but they can become the default answer for narrow ones that larger competitors never address.
The constraint is rarely whether an AI system has seen you. It is whether it can describe you with confidence.
Frequently Asked Questions
Can I pay to have AI assistants recommend my business?
Organic AI answers are not currently sold as placements. Some platforms are introducing advertising, but paid units are separate from the generated recommendations themselves.
How long does it take to start appearing in AI answers?
For search-grounded tools, changes to your pages and listings can show up within weeks. Changes to what a model "knows" without searching depend on future model updates and usually take much longer.
Does traditional SEO still matter?
Yes. Many AI search tools retrieve from search indexes, so being crawlable, indexed, and clearly structured remains a prerequisite for search-grounded answers.
Why does the AI describe my business incorrectly?
Usually because outdated or conflicting information exists somewhere it can access, such as old directory listings, outdated articles, or inconsistent profiles. Correcting the sources tends to correct the answers over time.
Key Takeaways
- AI systems don't place businesses in answers. They name businesses they can describe with confidence and find confirmed by independent sources.
- There are two pathways. Training data changes slowly, over months or years. Live retrieval through search-connected AI tools can change within weeks.
- The most common obstacle is not low visibility but inconsistent or vague descriptions of what the business does, for whom, and where.
- Third-party agreement from reviews, publications, directories, and community discussion often matters more than anything on your own website.
- Measure AI visibility as your share of mentions across many specific questions over time, not as a single yes-or-no check.
The Bigger Shift
Search visibility used to belong to individual pages: a URL ranked or it didn't. AI-assisted discovery makes visibility a property of the business itself, of how consistently and credibly it is described across the web. The unit of competition is shifting from the page to the reputation. The businesses most likely to be named in AI answers will not necessarily be the ones that publish the most. They will be the ones that are easiest to describe accurately in a single sentence, and whose description holds up when a machine checks it against everything else it can find.