By Steve Merrill, Founder of WRKNG Digital — November 2, 2026
How do you measure whether AI actually recommends your SaaS?
You build a set of 30 to 50 buyer prompts, run them through ChatGPT, Perplexity, Gemini, and Google AI Mode, then score three things: how often you get mentioned, how often you get cited, and where you rank when the answer lists options. Track those numbers monthly and you stop guessing.
Most SaaS teams I talk to track rankings and pipeline. They know their trial signups to the decimal. Ask them if ChatGPT names their product when a buyer types "best project management software for agencies" and you get a blank stare.
That blind spot is expensive. AI is now a top research surface for software buyers, and a big share of buying journeys start with a question typed into an assistant.
Why should SaaS marketers care about AI recommendations at all?
Because the buyer who asks an AI assistant "what are the alternatives to HubSpot" is deep in the evaluation stage. If your product doesn't show up in that answer, you were never in the running. No impression. No click. Nothing.
Google has rolled AI Mode into core search, with generative answers sitting at the top of results for a growing range of queries (Google, The Keyword). OpenAI pushed ChatGPT the same direction with built-in search that pulls live sources into answers (OpenAI).
Here's the thing. Rankings tell you where your page sits on a list of blue links. They say nothing about whether a model names you when a buyer asks a plain question. Two different games now, and buyers are playing the second one.
Step 1: How do you build a prompt set that mirrors real buyer questions?
Write down the questions your buyers actually ask, in their words, then turn each into a prompt. You want 30 to 50 that cover the main ways people research software in your category. Don't write prompts that only your product could win. That defeats the point.
Group them into three buckets:
- Category prompts. "Best X software for [audience]." "Top tools for [job to be done]."
- Alternative prompts. "Alternatives to [competitor]." "Cheaper options than [competitor]."
- Use-case prompts. "What should I use to [specific workflow]." "Software for [specific team] that needs [feature]."
Pull the real wording from sales call notes, support tickets, and win/loss interviews. The closer the prompt matches how a human talks, the more honest your results.
Step 2: How do you run the prompts across engines?
Run every prompt through the four engines your buyers use: ChatGPT, Perplexity, Gemini, and Google AI Mode. Paste the full answer into a sheet, and copy any sources the engine cites. Same prompt, same day, every engine. That keeps the comparison clean.
Answers vary by engine, so one tool naming you means nothing about the others. I've watched a client get named by Perplexity in almost every category prompt and skipped entirely by Gemini on the identical question. You only learn that by checking all four.
Run each prompt a couple of times too. These models aren't deterministic, so the same question can return a different list. Averaging a few runs smooths that out.
Step 3: How do you score mention rate, citation rate, and recommendation rank?
Score three numbers for every prompt. Mention rate is the share of answers that name your product. Citation rate is the share that link your site as a source. Recommendation rank is where you land when the answer gives a ranked or ordered list of tools.
Set it up as a simple grid. Prompt down the left, engine across the top, and in each cell mark yes/no for mention, yes/no for citation, and a number for rank if there's a list.
Then roll it up. If you're named in 12 of 40 category prompts on ChatGPT, that's a 30% mention rate. Write that down. It's your baseline. Ugly or not, you now have a real number where you had a guess.
Why three numbers instead of one? Because you can get named without getting cited, and cited without getting recommended. A model might link your blog post as a source while recommending a competitor in the actual answer. That gap tells you exactly what to fix.
Step 4: How do you track AI visibility over time?
Re-run the exact same prompt set on a fixed schedule. Monthly is plenty for most teams. Log every score next to last month's so you watch a trend line instead of reacting to a single scary snapshot.
AI answers move. Models get updated. Competitors publish new comparison pages. A third-party review site refreshes its rankings and suddenly you drop out of three answers. Without a schedule, you'd never catch it. With one, you see the dip the month it happens and you know where to look.
Keep the prompt set frozen. If you change the prompts every month, you can't compare the numbers. Add new prompts in a separate section and leave the core set alone.
Step 5: What do you fix when AI leaves your SaaS out?
When you're absent, it's almost always one of four reasons. The engine can't find a page that clearly answers the buyer's question, your comparison content is thin, trusted outside sites never mention you, or your pages lack the structure a model needs to parse them. Work through them in order.
Category pages. Build a real page for each category prompt you're losing. Not a feature dump. A page that directly answers "best X software for [audience]" with a clear, honest rundown.
Comparison pages. Buyers ask for alternatives constantly, so write head-to-head pages against the competitors they name. Be straight about where you win and where you don't. Models reward pages that read like a fair answer.
Third-party mentions. AI leans hard on sources it trusts. Review sites, industry roundups, Reddit threads, analyst posts. If those never mention you, the model has no outside signal to pull from. Earn those placements.
Structured data. Add schema to your product, pricing, and FAQ pages so engines can read what you offer without guessing. Google's documentation spells out the supported types (Google Search Central). It's the cheapest fix here and plenty of SaaS sites still skip it.
Fix the gaps tied to your highest-intent prompts first. Being named for "alternatives to [your biggest competitor]" beats winning ten vague prompts nobody buys from.
What does good look like once you've run this a few times?
Good looks like a mention rate that climbs month over month on your highest-intent prompts, a citation rate that shows engines trusting your pages, and a recommendation rank that moves up the list. You won't win every prompt. Win the ones buyers act on.
Data does not lie. Once you track these three numbers over time, every content decision gets easier.
Frequently asked questions
What is the difference between mention rate and citation rate?
Mention rate is how often an AI engine names your product in its answer. Citation rate is how often the engine links your site as a source. You can be mentioned without being cited, and cited without being recommended, so you track all three.
How many prompts do I need to measure AI visibility for my SaaS?
Start with 30 to 50 prompts that match how real buyers ask about your category. That's enough to spot patterns across engines without turning it into a full-time job.
Which AI engines should SaaS teams test?
ChatGPT, Perplexity, Gemini, and Google AI Mode cover most software research happening in AI today. Run the same prompt set across all four, because answers differ by engine.
Why does AI leave my SaaS product out of its answers?
Usually because the engine can't find a clear page that answers the buyer's question, or because trusted third-party sites never mention you. Fix your category pages, comparison pages, and earn outside mentions.
How often should I re-run the measurement?
Monthly works for most SaaS teams. AI answers shift as models update and as competitors publish, so a fixed schedule turns noise into a trend line you can act on.
Ready to find out if AI recommends your SaaS?
If you want help building the prompt set, running it across engines, and fixing the gaps, that's the work we do. See how we approach it at WRKNG Digital.
