Last updated: October 16, 2026
In a March to April 2026 survey of 519 US B2B professionals who use AI at work, published in July 2026, 92% said AI had shaped their vendor shortlist, and 71% said they use ChatGPT for product research. Only 7% said they noticed a vendor in an AI answer because they already knew the name.
Fit beats fame: 53% of the same respondents said they pay attention when a vendor closely matches their use case.
My verdict: for product prompts, ChatGPT mostly runs a live search and summarizes the third-party lists, comparison pages and vendor pages it finds. So the work is getting onto those lists, publishing honest comparison pages, and making your pricing and fit easy to quote. Schema tweaks, llms.txt and another round of top-of-funnel blog posts won't get you recommended.
This playbook covers the prompts that matter for B2B SaaS: "best [category] for [use case]", "[competitor] alternatives" and "[you] vs [competitor]". If you want the definitions first, read what AEO is and how it differs from SEO. This post assumes you're past that and want a to-do list.
How ChatGPT decides which products to name
There are two ways your product can end up in an answer, and they need different work.
The first is what the model already "knows" from training. Ask ChatGPT a broad question with search switched off and it answers from memory. That memory is built from text gathered before a cutoff date, so it lags reality by months. You can't edit it directly. You can only make sure the web says consistent, accurate things about you, so the next training run picks them up.
The second is live search, and for product prompts it's the one that matters most. OpenAI's help center says ChatGPT may search the web automatically when a question would benefit from current information, and that it rewrites your question into one or more targeted queries and sends them to search partners, naming Microsoft among them. Then it reads what comes back and writes a shortlist.
That means a "best X for Y" answer is often a summary of whatever pages ranked for a handful of rewritten queries a few seconds earlier. If those pages are third-party listicles that leave you out, you're out.
A few things follow from that:
Your product needs to appear on the pages that rank for the rewritten queries, not just on your own site.
Bing visibility matters more than most SaaS teams assume, because Microsoft is one of the providers ChatGPT queries.
Freshness matters. Pages that were updated recently get pulled more (more on that below).
ChatGPT has to be allowed to fetch you at all. If your robots.txt or CDN blocks its search crawler, you can't be cited in search answers.
What the studies say ChatGPT cites for software prompts
The research here is thinner than the confident LinkedIn posts suggest, and on review sites it disagrees.
An Ahrefs study of 750 top-of-funnel prompts, published in December 2025, found that "best X" blog lists were the most common page type among ChatGPT's sources, at 43.8% of all cited page types. Products that sat higher on those third-party lists were more likely to be recommended, and 79.1% of the 1,100 cited lists it dated had been updated in 2025, a quarter of them (26%) in the two months before the study. When a software brand's own site was cited, it was most often a landing page (37.2% of those prompts), not the homepage.
The same piece notes a separate Ahrefs finding that 28% of ChatGPT's most-cited pages have zero organic visibility in Google. A page-one Google ranking and a ChatGPT citation are related, but they aren't the same thing.
Review sites are where it gets contested. G2's own research, a March 2026 survey of 1,076 B2B software buyers, found 45% said citations from review sites were the most confidence-inspiring signal in an AI answer, and 64% said they run into inaccurate chatbot recommendations often or very often. The Semrush survey is cooler: after an AI names a vendor, only 38% go on to check G2 or similar sites, against 71% who visit the vendor's website. G2 sells review visibility, and Semrush sells AI-visibility tracking, so read both with that in mind.
My reading: multi-vendor pages (lists, comparisons, alternatives) carry the most weight, your own landing and pricing pages get cited more than most teams expect, and review sites matter more as a trust check after the answer than as the source of it. Treat all of it as directional; none of these studies is large, and the platforms change monthly.
Get onto the third-party lists ChatGPT reads
If I could only do one thing for a SaaS product that ChatGPT ignores, it would be this. Find the lists it's reading and get on them, ideally near the top.
Start by finding the actual pages. Run your 10 to 20 most important category prompts in ChatGPT with search on, open every citation, and paste the URLs into a sheet. Do the same for the rewritten versions you'd expect ("best [category] software 2026", "[category] tools for startups") in Bing and Google. Within an hour you'll have a list of 30 to 60 pages that decide your category.
Then sort them:
Lists that include you but rank you low or describe you wrongly. Email the author with a correction and something new: a feature launch, a pricing change, a customer number they can verify.
Lists that leave you out but include your direct competitors. These are your best targets. Offer the author a trial account, a short demo, or a clear one-paragraph summary of who you're for.
Lists on sites that look built only to collect affiliate fees or sell placements. Be careful here (see the mistakes section).
Two things make outreach work better than you'd expect. First, list authors update often, because fresh lists get cited more, so "you're missing a product" is useful news to them. Second, you're giving them something to write. A specific, sourced fact ("we added a native HubSpot sync in September") is far easier to add than "please include us."
What I'd skip: paying for placement on dozens of thin "top 10" sites. The Ahrefs study flagged that 35% of the cited lists it checked were on low-authority domains, many highly questionable, and it expects AI platforms to get stricter about them. Building on that sand is a short-term bet.
Publish comparison and alternatives pages on your own site
Because ChatGPT cites vendor pages so often, your own site is the one source you fully control. Three page types do most of the work for product prompts:
A "best [category] for [use case]" page that includes you and real competitors.
"[Competitor] alternatives" pages for the two or three tools buyers most often leave.
"[You] vs [competitor]" pages for the matchups that come up in sales calls.
I won't repeat how to build each one; there's a full guide to comparison, alternatives and category pages. The short version for ChatGPT specifically is this.
Should you rank yourself first on your own list? The Ahrefs study found 67.6% of the "best X software" Google results it checked had a list where the publisher ranked itself first, and recommended that SaaS companies publish them. Plenty of practitioners disagree on trust grounds. My view: put yourself first only where you can defend it for that use case, include competitors who genuinely fit, and say plainly who should pick someone else. A list that says "if you need X, choose Competitor B" is more believable, and buyers check.
Make each page answer a narrower question than your homepage does. The Semrush survey found 61% of buyers describe their specific use case when prompting, and 45% add constraints like budget or integrations. A page called "best [category] for 20-person agencies" matches that prompt better than "the best [category]."
Make your product facts easy to quote
When ChatGPT picks you, it has to say something about you. If your facts are vague, scattered or contradictory, it will either skip you or get you wrong.
Buyers notice when this goes wrong: 27% of Semrush's respondents said AI answers don't reflect real pricing or contract structures. Price, limits and fit are what they ask about. Yet a lot of SaaS pricing pages hide everything behind "Contact sales" or an interactive calculator.
Things worth stating in plain HTML on your own site:
Who the product is for, and who it isn't for.
Plan names, starting prices, billing period and what each plan includes. If pricing is quote-based, say what drives the quote.
Key integrations, regions and compliance claims, with dates where they change.
Limits that matter (seats, volumes, API rates).
A "last updated" date on pricing and comparison pages.
Then make the same facts match across the web. Your G2 and Capterra profiles, your LinkedIn page, your Crunchbase entry, partner directories and app marketplace listings should say the same thing about category, price range and ideal customer. When three sources say you're "for enterprise" and your site says "for startups", the model has to guess.
This is also where what GEO means in practice overlaps with this playbook: shaping the sentence ChatGPT writes about you, not just whether you appear.
Earn mentions on review sites and in communities
Review sites are where the evidence splits most, so I'd keep this proportionate.
The case for them: G2's buyer survey says review-site citations are what make an AI answer feel trustworthy, and review platforms' own comparison and category pages are multi-vendor pages of exactly the kind ChatGPT reads. You can only appear on those if you have a profile and enough reviews to be ranked. The case against over-investing: that finding comes from the review platform itself, and in Semrush's survey fewer than four in ten buyers check a review site after an AI recommendation.
My take: a complete, current G2 or Capterra profile with a steady trickle of genuine reviews is cheap insurance. Chasing badges and paid placements for the sake of ChatGPT isn't worth much until you've done the list and comparison work above.
Communities are similar. Reddit showed up in some studies and not others, and its weight differs a lot by engine. What's consistently true is that buyers verify: the Semrush survey found that after an AI names a vendor, 71% visit its website, 63% search for it on Google and 38% check reviews on G2 or similar sites. So genuine community presence helps twice, once as a possible source and once when the buyer goes looking.
Have your founders and product people answer real questions in the subreddits, Slack groups and forums your buyers use, with their names and affiliation visible. Don't seed fake "has anyone tried X?" threads. Moderators spot them, buyers resent them, and it's the kind of manipulation that ends up screenshotted.
Earned media is worth a line here too. Original data, a useful benchmark or a sharp opinion from your founder gives journalists and list authors a reason to mention you.
Fix technical access for OpenAI's crawlers
This is the boring part, and it's the one most likely to be silently broken.
OpenAI runs separate crawlers, and its crawler documentation is clear about what each does:
OAI-SearchBot surfaces sites in ChatGPT's search features. Sites that block it won't be shown in ChatGPT search answers, though they can still appear as navigational links.
GPTBot collects content that may be used to train OpenAI's models. Blocking it is a separate choice from blocking search.
ChatGPT-User fetches pages when a user's request triggers it, and OpenAI notes robots.txt rules may not apply because the action is user-initiated.
What to check this week:
Open yoursite.com/robots.txt and confirm OAI-SearchBot isn't disallowed, directly or through a blanket rule. OpenAI says changes take about 24 hours to take effect for search.
Ask whoever runs your CDN or firewall whether bot protection blocks OpenAI's published IP ranges. OpenAI's help center says your host or CDN needs to allow that traffic for the site to be eligible.
Make sure your key pages (pricing, comparisons, product pages) render their main content in the HTML, not only after JavaScript runs.
Check that those pages are indexed in Bing, using Bing Webmaster Tools.
You can block GPTBot and still allow OAI-SearchBot if you don't want your content used for training. For most B2B SaaS companies, I'd allow both. Training data is the slow path to being "known," and blocking it doesn't protect much.
What I'd treat as overrated here: schema markup and llms.txt. Both get sold as ChatGPT levers. The current evidence doesn't back that up, as the schema markup for AI search and llms.txt posts explain. Keep basic, accurate schema for Google. Don't expect it to get you recommended.
Test the prompts your buyers actually type
You can't improve what you only check once. ChatGPT's answers vary between runs, accounts and days, so a single screenshot proves nothing.
A simple test set for one product:
5 to 10 "best [category] for [use case]" prompts, using the use cases from your sales calls.
3 to 5 "[competitor] alternatives" prompts for your main competitors.
3 to 5 "[you] vs [competitor]" prompts.
2 or 3 prompts with constraints ("under $50 a seat", "that integrates with Salesforce", "for a team in the EU").
Run each one several times with search on, in a logged-out or clean session, and record:
Whether you were named, and in what position.
What it said about you, and whether that's accurate.
Which URLs it cited, and whether any of them are yours.
Which competitors were named instead.
The cited URLs are the gold. They tell you exactly which list to pitch, which comparison page to build, or which outdated fact to fix. Repeat monthly.
For a proper setup (prompt sampling, how much noise to expect, and how to tie it to pipeline), use the guide to measuring AI search visibility rather than reinventing it.
One warning on tools. Some trackers query the API, and API answers don't always match what users see in the app. That's fine for trends, but spot-check by hand before you act on a single result.
Mistakes to avoid
These are the patterns I'd steer you away from.
Writing more blog posts as the main fix. If ChatGPT's sources for your category are third-party lists, another top-of-funnel article on your blog changes little.
Generating hundreds of "[competitor] alternatives" pages from a template. It's the fastest route to thin pages, and Google's spam policies on scaled content apply whether a human or a model wrote them. Build the handful that match real buying conversations.
Buying placements on low-quality list sites. It may work for a few months, and then you're attached to sites the platforms are likely to downrank.
Faking reviews or community posts. Apart from the ethics, it's the sort of thing a competitor can expose in one Reddit thread.
Hiding pricing and limits. If ChatGPT can't find your price, it can't say you fit a budget constraint, and it may quote an old number from a third-party site instead.
Judging success by one prompt. Track a fixed set over time.
Treating ChatGPT as the only engine. In the Semrush survey, 61% of buyers also use Gemini for product research and 45% use Copilot. Do the shared foundations first, then tune per engine.
How this plays out in different SaaS categories
The playbook doesn't change much by category, but the weight on each step does.
In crowded categories such as sales engagement and cold email tools, the "best X" lists are numerous and updated often, so list outreach and alternatives pages usually come first. There are notes for sales enablement and cold email companies.
In categories with a more conservative buyer, such as accounting or recruiting software, review sites and consistent compliance and pricing facts tend to carry more weight. There's more on that for accounting software and recruiting software.
If you sell an AI or agentic product, your category name may not be settled yet, which adds a description problem on top of everything here. That's covered in the guide for AI and agentic AI companies and the AI company page.
One data point on what off-site and comparison work can add up to: Miraa's case study for the cold-outreach platform Salesforge, where the AEO/GEO and content work included competitor reviews, alternatives pages and pricing comparisons, reports 2,100 citations in Google AI Mode, 618 in Gemini and 388 in Perplexity (Ahrefs data, August 2026). It doesn't report a ChatGPT figure, so don't read it as proof of anything ChatGPT-specific.
Final thoughts
Getting recommended by ChatGPT is less mysterious than it looks. For product prompts, ChatGPT mostly searches, reads a small set of lists, comparison pages and vendor pages, and summarizes them. Your job is to be on those pages, accurately, and to make your own pages the easiest place to confirm the facts.
If I had to rank the work: get onto the lists ChatGPT already cites, build a few honest comparison and alternatives pages, publish clear pricing and fit, and check that OAI-SearchBot can reach you. Then test monthly. Skip the schema and llms.txt rituals until that's done.