Free guide · 2 Oct 2026
How to check whether AI assistants recommend your SaaS
"Does ChatGPT recommend us?" has no single answer: it depends on the question, the assistant, whether web search is on, and the day. Here's a method you can run by hand in about 30 minutes. It gives you a baseline you can trust more than a one-off screenshot.
- Write 10 questions in your buyers' words: category, problem, competitor and constraint questions.
- Ask each one in a fresh chat, in each assistant you care about. Note whether web search was used.
- For every answer, record: mentioned or not, position, how you're described, who's named instead, and which pages are cited.
- Ask every question at least twice, on different days, before drawing conclusions.
- Ask "What is [product] and how much does it cost?" separately, to check accuracy.
1. Write the questions your buyers actually ask
Don't ask "Is [my product] good?". Buyers rarely start there, and assistants will usually describe any product you name. The useful test is whether you come up when you aren't named. Mix four types:
| Type | Pattern | Example (invoicing tool) |
|---|---|---|
| Category | best [category] for [audience] | "What's the best invoicing software for freelancers?" |
| Problem | how do I [job] without [pain] | "How do I chase late invoices automatically?" |
| Competitor | [competitor] alternatives / [A] vs [B] | "What are good FreshBooks alternatives?" |
| Constraint | cheap / free plan / GDPR / for [stack] | "Which invoicing tool has a good free plan?" |
Use the words on your customers' sales calls and support tickets, not your tagline. If your homepage calls the product an "operating layer" but buyers say "tool", ask with "tool".
2. Ask in a clean setup
- Fresh chat, no custom instructions, no memory. Personalisation can change which products appear.
- Note whether the assistant searched the web. Answers that cite sources are built partly from pages fetched at answer time; answers without search lean on training data, which can be months old.
- One question per chat. Earlier turns influence later ones.
- By hand in the apps. Consumer apps generally don't allow automated scraping. If you want to automate, use the providers' official APIs (for example Perplexity's Sonar API or xAI's API with search enabled), and remember that API answers can differ from what the apps show.
3. Record five things per answer
A spreadsheet with one row per question per assistant is enough. Columns:
- Mentioned? Yes or no.
- Position in the list, if there is one. Being 6th of 7 and 1st of 3 are very different outcomes.
- Named instead: the products listed before you, or instead of you.
- Description: is the category right? Are the audience and price right? (See why assistants get pricing wrong.)
- Cited pages: copy the source links. This is the most actionable column.
4. How to read the results
Look at citations before positions. When an answer with web search cites the same three "best X tools" articles for most of your questions, those pages are shaping the answer. If you're not on them, that's a concrete gap you can work on: profiles, corrections, or asking the author to consider you. If your own pages are cited but you're still not named, the page probably doesn't say plainly enough what the product is and who it's for.
Phrasing matters more than people expect. In one of our own checks, a form-backend tool appeared at position 3, 4 and 6 in three slightly different phrasings of essentially the same question (Perplexity Sonar API, web search on, 2 Oct 2026). Test several phrasings before deciding whether you "rank".
One answer is a snapshot, not a ranking. In our own runs, a product listed 5th for a question on 1 Oct 2026 was missing entirely when we asked the same question on 2 Oct (Perplexity Sonar API, web search on). Ask each question at least twice, on different days. Treat "mentioned in most runs" and "never mentioned" as the signals.
5. Check accuracy separately
Ask each assistant "What is [product] ([domain]) and how much does it cost?". You're looking for a wrong category, outdated plans, or prices that don't match your pricing page. These are often easier to fix than visibility, because the cause is usually a specific page you can find in the citations.
6. Repeat monthly
Keep the same questions and re-run them every month. Changes to third-party pages and to crawler access can take weeks to show up. A stable question set is what makes the comparison meaningful.
Want this automated?
Our AI Visibility Report ($49 PDF) is generated by software that asks Perplexity and Grok 10 buyer questions through their official APIs with web search on. It records mentions, positions, products named instead, cited pages, how you're described, and quoted prices compared with your homepage. It doesn't cover ChatGPT, Claude or Gemini yet. Or request a free mini report (3 questions, Perplexity).
More guides
- How to get your SaaS recommended by AI assistants: a checklist
- Why AI assistants get your SaaS pricing wrong
- Can AI crawlers read your homepage? A 5-minute JavaScript check
Sources
- Perplexity API documentation (Sonar): docs.perplexity.ai
- xAI documentation, search in the API: docs.x.ai
- OpenAI, Overview of OpenAI Crawlers: developers.openai.com/api/docs/bots
- Observations marked "our own checks" come from single API runs we made on 1–2 Oct 2026; they are examples, not statistics.
Independent guide; not affiliated with OpenAI, Anthropic, Google, Perplexity or xAI. Product names are trademarks of their owners.