You are about to buy something for your business, so you open ChatGPT and type “best [your category] for a mid-size team” before you talk to a single salesperson. Three names come back with a sentence of reasoning each. You click one, maybe two. The other twenty companies in that category, including some that are better than the three you were shown, never enter your consideration set. They were not rejected. They were never surfaced. This is the new top of the funnel, and getting AI chatbots to recommend your service is now the same kind of fight that ranking on page one of Google used to be, except the page only has three results and there is no scrolling to position eleven.
The instinct most founders have is to write a better page about why they are the best. That instinct is wrong, and understanding why is the whole game. When a model answers “who should I hire for X,” it is not reading your homepage and being persuaded. It is assembling a consensus from the sources it trusts on that question, then naming the companies those sources name. Your marketing copy is not a trusted source about whether you are good. Other people talking about you is. So the work is not persuasion, it is becoming the name that credible third parties keep repeating, in a form the model can extract. Here are seven moves that do that.
Move one: map the recommendation surface before you touch it
Before you change anything, find out what the models already say. Open ChatGPT, Perplexity, Claude, and Gemini, and ask each the five or six questions a real buyer in your category would ask. Not “is [your company] good,” which biases the answer, but the open questions: “best options for,” “alternatives to [the market leader],” “who specializes in [your niche].” Write down exactly which companies get named, in what order, and which sources the answer leans on when the platform shows them.
I call the set of questions where you could plausibly be recommended your recommendation surface, and most companies have never looked at theirs. When you do, two things become obvious. First, the same five or six competitors tend to dominate across every model, because they share the same underlying source base. Second, the answer almost always traces back to a small number of third-party pages: a roundup article, a review platform, a Reddit thread, a comparison post. Those recurring sources are your targets. You are not trying to win everywhere; you are trying to get into the handful of places the model already reaches for when it answers your buyer’s question.
Move two: get named in the third-party roundups the model already cites
The single most powerful action is getting your company added to the “best [category]” and “top [category] tools” articles that the model cites when it builds a recommendation. These roundups are where AI chatbots recommend service providers from, because they are exactly the kind of structured, comparative, third-party content models trust for a shortlist question. If three of the four sources behind an answer are listicles you are absent from, you will not be named, no matter how good your own site is.
Finding them is the work you already did in move one. Getting in is outreach: contact the authors and editors of those pieces with a specific, useful reason to include you, ideally a real differentiator and a concrete proof point, not a generic “please add us.” Some will update for free because keeping a roundup current serves their readers. Some run on a paid-inclusion or affiliate model, which is a legitimate distribution cost. Either way, every roundup you join is a source the model can pull your name from, and because these articles get refreshed and re-cited, the value compounds rather than decaying like a one-time mention.
Move three: build the entity so the model knows what you are
A model can only recommend you for a query if it confidently understands what you do, who you serve, and how you differ. That confidence comes from entity clarity, the consistent, structured definition of your business across your own site and the wider web. When your category, your specialty, and your ideal customer are stated the same way everywhere, the model can match you to the right question. When they are muddy or contradictory, the model hedges and names someone clearer instead.
On your own site, this means unambiguous service pages that state plainly what you do and for whom, backed by Organization and Service structured data so machines read the facts without guessing. Off your site, it means your descriptions on review platforms, directories, and your own social profiles agree with each other. The goal is that a model assembling an answer about your category finds a single coherent entity, you, defined the same way in ten places, and can therefore slot you into the exact query you should win rather than leaving you out because it was not sure where you fit.
Move four: win the review platforms that feed the answer
When buyers ask AI chatbots to recommend a service, the models lean heavily on review platforms, because reviews are the closest thing to a trusted, quantified verdict from real customers. A strong, recent, specific review presence on the platforms that matter in your category does double work: it raises the odds you are named, and it shapes the one-sentence reason the model gives for naming you. Models often paraphrase review themes directly, so the language your customers use becomes the language the model uses to recommend you.
This is not about gaming a star rating. It is about a steady flow of genuine reviews that mention specific outcomes, use cases, and the kind of customer you serve. A review that says “great for ecommerce teams migrating off spreadsheets” gives the model a precise hook to recommend you for exactly that query. Thin, generic, or stale reviews give it nothing to work with. Build the habit of asking happy customers to review you in the places your buyers and the models both look, and ask them to be specific about what you actually solved.
Move five: be present where your buyers argue in public
Models draw heavily on community discussion, Reddit threads, niche forums, and Q&A sites, because those places contain candid, real-world opinion that reads as more trustworthy than marketing. When someone in your category asks “what does everyone use for X” and a thread fills with genuine recommendations, that thread becomes a source models cite, sometimes verbatim. If your company never appears in those conversations, you are absent from a major input to the answer.
You cannot fake your way into this, and trying gets you caught and burned. What works is genuine participation: customers who like you mentioning you unprompted, your own team showing up to answer questions honestly without a hard sell, and you being good enough that people bring you up on their own. The communities that matter in your niche are findable, and being a real, helpful presence there over time seeds the kind of organic mentions that models treat as credible signal. It is slower than buying an ad and far more durable, because it shapes the consensus rather than interrupting it.
Move six: publish the comparison content buyers actually search
Buyers asking a model for a recommendation are often really asking a comparison question: you versus the market leader, you versus the obvious alternative. Models answer those with comparison content, and most of that content is written by third parties or by your competitors, which means the framing rarely favors you. Publishing your own honest, specific comparison material, and earning third-party comparison coverage, puts accurate information about your differences into the source base the model reads.
Honest is the operative word. A comparison that pretends you win on every axis reads as marketing and gets discounted. One that says plainly where you are the right choice and where a competitor fits better is more credible to both humans and models, and it gets cited because it is genuinely useful for the decision. The aim is to make sure that when a model weighs you against an alternative, it is working from a fair, specific account of your real strengths rather than from a competitor’s framing of your weaknesses.
Move seven: make every page extractable, not just readable
The last move is mechanical and easy to skip. Models pull recommendations from content they can parse cleanly, so the structure of your pages affects whether your facts survive the trip into an answer. A claim buried in a long paragraph of clever copy is harder to extract than the same claim stated plainly under a clear heading. Specific, self-contained statements travel; vague, decorative ones do not.
Practically, that means clear headings that match real questions, direct answers placed right under them, specific numbers and named use cases instead of adjectives, and structured data that hands the facts to machines without interpretation. You are writing for two readers at once: the human who needs to be convinced and the model that needs to be able to lift a clean, true sentence about you into its answer. Do all seven moves together and the pattern holds: you stop trying to be the most persuasive voice in the room and start becoming the name that every trusted source in your category already agrees on, which is the only thing a model can actually repeat.
The mistake that undoes all seven moves
There is one shortcut that tempts every founder who understands how this works, and it backfires every time. Once you grasp that models build recommendations from third-party consensus, the obvious hack is to manufacture that consensus: fake reviews, planted forum posts, thin sites that exist only to name you, coordinated mentions that look organic but are not. It works for a brief window and then it does not, because the platforms, the review sites, and the communities all actively hunt for exactly this pattern, and the models increasingly weight sources by how trustworthy they are rather than how numerous.
When manufactured signal gets caught, and it does, the cost is not neutral. Reviews get purged, posts get removed, and the communities that matter remember the brand that tried to game them. You do not just lose the fake signal; you poison the real consensus, because now there is a visible record of you trying to manipulate it, and that record is itself a source the model can weigh against you. The downside is asymmetric: a slow honest base compounds quietly in your favor, while a fast fake one can detonate the moment it is exposed.
The reason the seven moves work is that every one of them changes the genuine consensus rather than faking it. Getting into real roundups, earning real reviews, participating honestly in real communities, publishing honest comparisons, these all make the true picture of you more visible and more extractable. That is durable because it is real, and it is real because you earned it. The model is summarizing what credible people actually think of you, so the only reliable way to change the summary is to be worth a better one and then make sure the evidence of it is everywhere the model looks. Do the slow version. It is the only one that holds.