"Which project management tool should I use for a ten-person agency?" gets a different answer, built from different source material, than "what is Asana." The first is a recommendation prompt. The second is a citation prompt. Most of the advice floating around about ranking in ChatGPT treats these as the same problem, and they're not.

Our piece on getting cited by ChatGPT and Google AI Overviews covers the citation case: how a single page gets pulled and attributed for a specific claim. This one is about the harder, more commercially valuable case: getting named as the actual recommendation when someone asks ChatGPT to pick something for them. That's a different mechanism, pulling from different source material, and optimizing for citation alone won't get you there.

Why a recommendation prompt behaves differently

A citation prompt needs one good page. ChatGPT retrieves a candidate set, finds the passage that answers the question cleanly, and attributes it. A recommendation prompt asks the model to make a judgment call among several options it already has some opinion about, formed from a much wider pool of material than whatever it retrieves in that one session. The model isn't just looking for a page that defines your product. It's synthesizing across everything it has encountered about your category: comparison articles, reviews, forum threads, your own site, your competitors' sites, and whatever baseline impression got baked in during training.

That's why a product with a single, well-optimized landing page can still get skipped in a "best X" answer: the page answers "what is this" perfectly well and still doesn't move the needle on "should this be recommended over the other four options," because that judgment draws on corroboration across sources, not extractability from one source. The work in this post is mostly about building that corroboration deliberately, on top of the retrieval basics you need regardless.

None of this works if the fundamentals aren't in place first. Your site still needs to be crawlable, indexed, and structurally sound, and the robots.txt and crawler-access points covered in our citation guide still apply here exactly as written. If you're new to this category generally, GEO for beginners is a reasonable starting point before this post. What follows assumes the fundamentals part is already handled.

Get into the roundups you're not writing yourself

The single most effective thing in this category is also the hardest to fully control: showing up in comparison and "best X" content published by someone other than you. A roundup article ranking the top project management tools, a Reddit thread debating two competitors, a review site's head-to-head comparison, all of these feed the same corroborating-source pool a recommendation answer draws from, and they carry more weight than your own marketing copy saying the same thing, because the model (like a human reader) treats third-party agreement as a stronger signal than self-description.

You can't force an independent publication to include you. You can make inclusion easy and likely. Build a straightforward, genuinely honest comparison page on your own site for each real competitor your buyers actually bring up by name, not a strawman matchup where you win every row. That page does two jobs: it's citable on its own, and it hands outside writers a factual skeleton (feature lists, pricing tiers, the actual differences) to work from when they build their own roundup, which is often exactly where that content gets pulled from.

Pitch inclusion in existing roundups directly rather than waiting to be noticed. Identify the three to five "best [category] tools" articles your real competitors already appear in, and reach out with a specific, useful pitch: what makes your tool a fair fit for that list, and what you'd want said about it that's actually true. This is ordinary PR and SEO outreach, not a new discipline invented for AI search, and it still works for the same reason it always did: a credible third-party site with real traffic and real authority gets crawled, gets cited, and gets fed into retraining and retrieval far more reliably than most individual company blogs ever will.

Reviews matter here differently than they do for citation

A product's review profile on G2, Capterra, TrustRadius, or Trustpilot rarely gets cited as a single source the way a how-to article does. It feeds recommendation judgments in aggregate instead. A product with dozens of specific, recent reviews naming concrete features and concrete complaints gives a model actual material to draw a nuanced recommendation from ("good for small teams but support response times are a common complaint"). A product with three generic five-star reviews and nothing else gives it almost nothing, and models tend to hedge or simply omit a brand they can't say anything specific about, rather than recommend it on faith.

If you're not actively asking recent, satisfied customers to leave a specific review on the platform your actual buyers check (G2 skews B2B software, Capterra and TrustRadius overlap with it, Trustpilot and industry-specific sites matter more elsewhere), this is the most mechanical, controllable lever in this whole post. It's also slow. Review volume and specificity build over months, not days, so starting now matters more than getting the request wording perfect. Our broader piece on earning AI mentions goes into more depth on review platforms and other off-site signals if this is the first time you're building this muscle.

Entity consistency: make it easy to know you're one thing

Models forming a recommendation have to resolve, implicitly, whether the "Acme" mentioned in a 2024 review, the "Acme Software" on your homepage, and the "Acme Inc." in a press release are the same entity. Inconsistent naming across your own properties makes that resolution harder, and a model that's uncertain whether sources are even talking about the same product tends not to feature that product confidently in a recommendation, because the confidence a recommendation requires depends on the sources agreeing about what they're describing.

Use one consistent name for your product and company across your site, your social profiles, your review platform listings, and your schema markup. Back it with Organization and Product structured data that matches what's visibly on the page, since Google's structured data guidelines are explicit that markup has to reflect the visible content, not aspirational or inconsistent branding (Google structured data general guidelines). If your company has gone through a rebrand or a name change, audit your own site and your most-cited third-party listings for the old name still lingering, since that split identity actively works against you here in a way it wouldn't for a simple citation.

The Bing question, and what's actually confirmed

This one needs care, because the common advice ("get indexed in Bing because ChatGPT search runs on Bing") is based on something that was true and is getting less true. When OpenAI launched ChatGPT's search feature in October 2024, it built the live retrieval layer on top of Microsoft's Bing index, a natural extension of Microsoft's investment in and technical partnership with OpenAI (CNBC). OpenAI's own current crawler documentation describes a dedicated crawler, OAI-SearchBot, built specifically to "surface websites in search results in ChatGPT's search features," and states plainly that "sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers" (OpenAI developer docs). That page doesn't mention Bing at all, which is itself informative: OpenAI's documented, current description of how ChatGPT search retrieves content centers on its own crawler, not a third-party index.

What we can't confirm with a primary source is the exact current split between Bing-sourced and OpenAI's-own-index-sourced results for the consumer ChatGPT product, and reporting through 2026 describes that balance as having shifted substantially toward OpenAI's own infrastructure since the 2024 launch. The practical takeaway that holds regardless of the exact split: confirm OAI-SearchBot isn't blocked in your robots.txt, since OpenAI states directly that this determines ChatGPT search eligibility, and treat Bing indexing as cheap, additional insurance rather than the primary lever. Submitting your site through Bing Webmaster Tools costs nothing and also helps your visibility in Microsoft Copilot, which Microsoft has confirmed runs on Bing's index with no separate Copilot-specific crawler, so that effort isn't wasted even if ChatGPT's own reliance on Bing has genuinely declined.

OAI-SearchBot access is necessary, and nowhere near sufficient

Checking robots.txt for OAI-SearchBot access is a five-minute task and most sites that bother to check pass it without changing anything, because the accidental-blanket-block problem mostly affected GPTBot (added during 2023's training-data panic) rather than the separately-controlled search crawler. Confirm it anyway; a site that can't be crawled for search purposes is not eligible for a citation-based mention, let alone a recommendation, no matter how good its reviews or roundup coverage are.

But crawler access gets you into the candidate pool. It does nothing to make the model confident enough to recommend you over a competitor it's equally able to crawl. That confidence comes from everything covered above: the roundups, the reviews, the consistent entity signals. Treat the robots.txt check as a gate you clear once and move past, not a lever you keep pulling expecting different results.

Build the pages that answer the comparison question directly

Beyond the competitor comparison pages already mentioned, build and maintain a current pricing page that actually states your pricing, in plain terms, without requiring a sales call to find out. "Contact us for pricing" is a dead end for a model trying to answer "is X affordable for a small team," and a dead end for the human reader asking the same thing. If your pricing genuinely can't be public (enterprise, highly custom), at minimum publish a representative range or a starting price, since a model forced to guess at affordability from no data at all tends to default to recommending whichever competitor made that information available.

Build a page that directly answers "is [product] good for [specific use case]" for your two or three most common use cases, in the actual language buyers use to ask that question, not generic feature-list marketing copy. This is the same answer-first, self-contained-fact structure that works for citation, applied specifically to the comparison and suitability questions a recommendation prompt is actually trying to resolve. A model trying to decide whether to recommend you for "a ten-person agency" has an easier job if a page on your own site explicitly addresses that scenario than if it has to infer suitability from a generic features page built for every customer size at once.

Keep these pages current. A recommendation prompt about pricing or fit is exactly the kind of query likely to trigger a live search rather than rely on stale training data, since it's asking for something that genuinely changes over time, so an outdated price or a feature list that hasn't caught up with your product for a year works against you more here than it would on a purely informational page.

A worked comparison page, briefly

Here's the shape that tends to work versus the shape that doesn't, since "write a comparison page" is advice vague enough to produce a lot of bad pages. A weak version opens with "Both Acme and Competitor are popular choices for teams looking to manage projects effectively" and spends three paragraphs on shared history before getting to any actual difference. That's not wrong, exactly, it's just not useful to anyone trying to decide, human or model.

A page that actually earns its place states the real difference first: "Acme charges per seat starting at $12/month with unlimited projects on every tier; Competitor charges per project with seats bundled, which works out cheaper for a 3-person team and more expensive past about 15 people." Then it backs that claim with the specifics: a short table of actual plan tiers, the one or two features each product handles differently, and an honest note on what the competitor does better, because a comparison page that never concedes a single point reads as marketing to a model the same way it reads that way to a person, and a model weighing source credibility has reason to discount a one-sided comparison more heavily than a balanced one.

Track whether any of this is actually moving the needle

Everything above takes weeks to show up, not days: review volume builds slowly, a roundup pitch might not land for a month, entity consistency fixes need time to get re-crawled and re-absorbed. Don't wait until a quarterly review to find out whether it worked. Run the same recommendation-style prompts on a schedule using the manual testing method in our guide to checking whether ChatGPT mentions your brand, specifically the "best X for Y" and "which should I use" prompt shapes rather than direct brand-name queries, since those are the ones this entire post is trying to move. A rising mention rate on recommendation-style prompts, specifically, is a better signal that this work is paying off than a rising mention rate on direct brand queries, which tends to move for unrelated reasons (brand awareness, press coverage) that have nothing to do with roundups or review volume.

For ecommerce brands, recommendation prompts often turn into shopping results, so pair these tactics with our ChatGPT merchant setup guide. And for new AI startups, the press and directory coverage that comes with joining one of the AI accelerators backing AI founders is often the first set of third-party mentions ChatGPT can find.

Google's answer engines are moving the same way. Our breakdown of Google's Universal Commerce Protocol shows how AI Mode and Gemini can go from recommending a product to buying it for the shopper.

To see whether these tactics are working, track a fixed set of recommendation prompts over time. Our list of ChatGPT rank tracker tools shows which ones handle that well.

Organic recommendations aren't the only way into ChatGPT anymore. Our breakdown of ChatGPT ads covers OpenAI's new ad formats and whether they're worth testing.

FAQ

Is getting recommended by ChatGPT the same as getting cited?

No. A citation attaches a link to a specific claim pulled from one retrieved page. A recommendation is a judgment call the model makes across a wider pool of source material, including things you don't control directly like reviews and third-party comparisons. You can get cited without ever being recommended, and in principle get named in a recommendation with no citation or link attached at all.

Does ChatGPT search still rely on Bing?

It launched on Bing's index in October 2024. OpenAI's current, documented description of ChatGPT search retrieval centers on its own OAI-SearchBot crawler, with no mention of Bing in that documentation, and reporting through 2026 describes OpenAI's reliance on Bing as having declined since launch, though the exact current split isn't something either company has published precisely. Confirming OAI-SearchBot access in robots.txt is the documented requirement; treating Bing indexing as a secondary, low-cost precaution (and one that also helps Copilot visibility) is the safer bet than either ignoring Bing entirely or treating it as the primary lever. Submitting a sitemap through Bing Webmaster Tools takes a few minutes, and Microsoft shipped its own first-party AI Performance report inside that same tool in February 2026, showing site owners Copilot citation counts and the actual grounding queries that produced them, which is more transparency into a Bing-powered AI surface than most of this category offers today (Bing Webmaster Blog).

How much does review volume on sites like G2 actually matter for a ChatGPT recommendation?

It's one of the stronger off-site signals in this category, though nobody including OpenAI publishes a precise weighting. The logical mechanism is straightforward: a model trying to recommend something specific and defensible needs specific material to draw on, and a thin or generic review profile gives it almost nothing to work with for your product compared to a well-reviewed competitor.

Should I rewrite my pricing page if we don't publish exact prices?

Publish what you reasonably can. Even a representative starting price or range beats "contact sales" for both the model and the human reader trying to answer a basic affordability question, and a total information vacuum tends to work against you when a competitor has made the same information available.

Can I get recommended without appearing in any third-party roundup at all?

It's harder, not impossible. A product with an unusually strong, specific, and dense review profile and clean entity signals can still surface in a recommendation without ever being the subject of outside comparison content, but roundups and comparisons remain the one most companies in this position can actually go out and build on purpose, since review volume accumulates slowly and roundup inclusion can be pursued directly through outreach this month.

Pick the three competitors your sales team hears named most often in deals, right now, and check whether you have an honest comparison page against each one live on your own site. If you don't, that's the single most direct piece of this whole post to fix first, and it's the one most directly inside your control.