Getting recommended by the AI engines is 2 jobs: earn the third-party evidence they trust, and make your own site the source they quote. Here's the process for B2B SaaS, with the data behind each step.
- Pick the buying prompts you want to own, sourced from real sales conversations, weighted to the decision stage.
- Build the evidence layer: reviews on the platform AI cites in your category, listicle placements, independent comparisons. Third-party comparisons get cited at 3x the rate of vendor us-vs-them pages.
- Make your own pages quotable: ChatGPT pulls 74.6% of product-query citations straight from vendor sites.
- Track it monthly with 4 tiers: Recommended, Cited, Mentioned, Absent.
Ask around about getting recommended by ChatGPT and most of what comes back was written for local businesses: claim your Google Business Profile, collect Yelp reviews, add FAQ markup. Useful for a dentist, close to useless for a B2B SaaS. This is the SaaS version, built from our citation study and the client work that came out of it, and it splits into 2 jobs you run in parallel.
How the engines decide who to recommend
The engines compose answers from 2 layers. On Perplexity, Gemini, and Claude, 79.0% of product-query citations point to third-party sources: the reviews, listicles, and comparisons written about you. ChatGPT flipped that pattern and now pulls 74.6% of its product-query citations directly from vendors’ own sites. So the work splits into 2 jobs, and skipping either one caps the whole effort. It’s just how the game works now. If it is Perplexity in particular where you are missing, the causes run in a different order, and we have set those out in why your B2B SaaS doesn’t show up in Perplexity.
Nothing about that split is random. When a buyer asks “best CRM for a 50-person SaaS,” the engine wants corroboration, so it reads what independent sources say. When the same buyer asks about your pricing, ChatGPT goes straight to your pricing page with a site: search. Each job below serves one of those retrieval habits, and the full breakdown is in our citation study.
Decide which prompts you want to own
Before touching content, write down the 30 to 50 questions a buyer with budget would ask an engine about your category: the “best X for Y”, “X vs Y”, and “is X worth it” phrasings. We tend to pull these straight from sales-call transcripts, because buyers’ words beat keyword-tool words, and the transcripts also tell you which competitors and constraints buyers name unprompted.
Weight the set toward the decision stage, and split it 3 ways: prompts you must win, prompts where you’re fighting named competitors, and experimental prompts nobody has claimed yet. A buyer asking a definitional question is reading; a buyer asking which product fits their situation is choosing, and the choosing prompts are where recommendations get made. This prompt set becomes both your to-do list and your scoreboard.
Build the evidence layer
This is the slower job and the bigger one. 3 source types carry most of the weight in our citation data: reviews on the platform AI already cites, listicle placements at the consideration stage, and independent comparisons at evaluation. Each one responds to deliberate work, and each feeds a different buying moment.
Reviews come first, but on the right platform. Find which review site the engines cite for your category by reading your own AI answers, then concentrate there rather than spreading thin. Brief reviewers to name the features they use and the problems those features solve, because a review that says “great product” is a rating, while a review that names a feature is a sentence an engine can quote.
Listicles decide the consideration stage. In our data, half of all consideration-stage citations come from “best X for Y” roundups. Earn placements in the ones that exist, write your own that includes competitors on their merits, and treat the roundups nobody has written for your niche as open territory. Vendors listing themselves alongside rivals feels cheeky, and it works: the engines pick those lists up like any other.
Independent comparisons decide the evaluation stage, and they carry 3x the citation weight of vendor-written us-vs-them pages. You can’t write “independent” yourself, which is where digital PR earns its keep: getting credible third parties to compare you is the single strongest evidence play in the set.
Make your own site the source
ChatGPT’s shift to vendor sites means your own pages are back on the front line. It runs searches like site:yourdomain.com pricing while composing answers, so the pages it lands on need to answer like a person would: real numbers, clear feature names, current information, and tables rather than images, because the engines read text and skip pictures.
The bar for quotable is specific. In our data, 80% of decision-stage citations point at pages containing concrete figures, so a pricing page that says “contact us” gives an engine nothing to repeat, and it moves on to one that quotes a number.
On schema and llms.txt: we feel as though both get oversold, and we treat them as nice add-ons rather than the game. The engines read well-structured pages fine without markup, and no tag rescues a page with nothing quotable on it. Put the hours into pages with real answers and the evidence layer above.
Show up where the engines listen
Reddit sits near the top of ChatGPT’s consulted sources while barely appearing in its visible citations, and community threads shape answers on every engine. The rule for showing up there without torching your reputation: be useful first, and mention your product only when it answers the question asked.
Let your team participate as named people rather than a brand account. One genuinely helpful answer in the right thread outworks 50 promotional ones, and the promotional ones get you banned anyway.
LinkedIn plays a similar role for B2B categories: the engines read it, buyers research on it, and consistent expert commentary there becomes part of the corroboration trail an engine follows when it decides if it can trust you.
Track whether it’s working
Run your prompt set monthly, on the same engines, the same way, and classify every appearance into 4 tiers: Recommended, Cited, Mentioned, Absent. Watch 3 numbers over time: your share of voice against named competitors, your recommendation rate on the prompts that matter, and the accuracy of what the engines say about you.
Add a how-did-you-hear-about-us field to your demo form too, because buyers who type “ChatGPT” into it are attribution no analytics tool will ever give you.
I’d say the tier ladder is the scoreboard that can’t be gamed. Moving from Absent to Mentioned means the evidence is landing; Mentioned to Recommended means the engines now trust it. If you want the full measurement process, the audit guide covers it step by step, spreadsheet included.
Frequently asked questions
The 4 questions teams ask us most about getting recommended, answered the way we answer them on calls: how long it takes, whether you can pay for it, what markup does, and whether the engines need different strategies.
How long does it take to get recommended by ChatGPT?
Weeks for your own-site fixes to register, months for the evidence layer to compound. In our client work, meaningful tier movement showed inside a quarter, with recommendation rates climbing from there. Anyone promising placement in days is selling something the engines don’t offer.
Can you pay to appear in ChatGPT answers?
No. There’s no ad unit inside the answers and no submission form. It obviously comes down to what the engines read, and nobody sells placement in that. Every appearance is earned through what the engines read: your pages and other people’s. That’s uncomfortable and it’s also the moat, because a recommendation that can’t be bought is one buyers trust.
Does FAQ schema get you recommended?
We don’t rely on it. Markup neither blocks nor guarantees anything; the engines quote pages that answer questions plainly, with or without tags. Write the answer first. If your platform adds schema automatically, fine, but don’t spend strategy hours there.
Do ChatGPT and Perplexity need different strategies?
Different weightings of the same strategy. Perplexity leans on third-party content, so evidence work pays more there. ChatGPT leans on your own site, so page clarity and crawler access pay more. Run both jobs and every engine improves at once.
