How to get your SaaS recommended by ChatGPT

Step 1: write down the questions, not the keywords

Keywords are what people type into Google. Questions are what they ask an assistant. They are longer, they carry a situation ("for a remote team of 12"), and they usually ask for a shortlist.

Start with 20 to 50 of them. Pull them from sales calls, support tickets and the "People also ask" box. Then sort them into three buckets: category ("best help desk software for SaaS"), comparison ("X vs Y", "alternatives to Y") and use case ("help desk with a shared inbox and Slack"). Comparison questions are where a buyer is already holding two tabs open, so they matter most.

Step 2: get a baseline, and run it more than once

Ask each question in ChatGPT with web search on, in Perplexity and in Gemini. Write down three things for every answer: are you named, in what position, and which URLs are cited.

Do it again later. AI answers are sampled, not looked up. In our own runs on 10 Nov 2026 we ran the same 10 questions three times in one day for the same company. It was named in 5 of 10 questions, then 5, then 2. One run is an anecdote. A tracker that repeats the questions daily is the only way to see a trend through that noise. Our free audit runs one pass and says so on the report.

Step 3: read the citations, not the answer

The answer is the symptom. The cited pages are the cause. When ChatGPT and Perplexity use search, they build the answer from a handful of pages, and they show you which ones.

Collect those URLs across all your questions and group them. In the runs behind our category research, the same few kinds of page show up again and again: review and comparison sites such as G2, Capterra and SelectHub, big "best X" listicles from publishers, and the blogs of the largest vendors in the category. If your name is not on those pages, the model has little to name.

Step 4: fix the pages the AI reads

There are four places to work, in this order. All of them are legitimate, and all of them are slow.

1. Your own site. Publish the pages that answer comparison questions directly: "[You] vs [Competitor]", "[Competitor] alternatives", "best [category] for [use case]". Be fair to the other product. Pages that read like honest evaluations get cited. Pages that read like ads do not. Put real prices and real limits on them.
2. The listicles the AI already cites. Find the publishers behind the URLs from step 3 and ask to be added, with a factual summary and something they can verify. A short, polite email beats a bought placement every time, and bought placements are the thing that gets a site removed from AI answers later.
3. Reviews from real customers. G2, Capterra and Trustpilot show up in citations constantly. Ask customers who are happy to say so. Never buy, write or trade reviews. That is against the rules of those sites and against FTC guidance, and it does not last.
4. Communities, under real names. Reddit, LinkedIn and X answers by your real team, with the affiliation stated, can be cited. Fake accounts get removed, and when the removal happens, the citation goes with it.

Step 5: make it easy for crawlers

None of this matters if the page cannot be fetched. Check these once and move on:

• robots.txt allows GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot and Google-Extended (and that you actually want them to).
• The pages are server-rendered. A crawler that does not run JavaScript should still see your comparison table.
• Structured data: Organization, Product or SoftwareApplication, FAQPage where you have real questions and answers.
• A plain /llms.txt that says in a few lines what you are and links your best pages. No major engine has promised to use it, so treat it as cheap and optional. Ours is here.
• Fresh dates. Say when the page was last updated, and update it.

Step 6: measure for 90 days before you judge it

New pages typically get picked up in weeks, not days. Third-party placements and reviews take longer. Track the same questions daily and look at three numbers: the share of questions where you are named, your position when you are named, and the number of cited pages that mention you.

Be careful with one claim. A page being cited is not the same as the AI recommending you. Cited means it read you. Recommended means it chose you. Track both.

What not to do

• Do not buy aged Reddit accounts, upvotes or comments. Reddit's citation share in AI answers has moved a lot (DesignRush, citing Semrush, reported a drop from roughly 60% to roughly 10% between August and mid-September 2025), so a plan that depends on it is fragile even before the bans.
• Do not stuff a page with the brand name or hide text for crawlers. It is the old spam, and it reads as spam to a model too.
• Do not promise yourself a ranking. Nobody controls these answers. You can only become easier to cite.