Short answer. To check if AI recommends your software product, ask ChatGPT, Claude, Gemini and Perplexity the questions a buyer would type, such as "what is the best invoicing tool for a five-person agency?", with web search on and in a private window. Record whether your product is named, whether your website is linked, how it is described, and which competitors appear.
Why software is a different case
A software buyer often starts with a problem and asks an AI for a shortlist, then compares the names on it. If your product is missing from the shortlist, it is missing from the comparison. And because software changes quickly, an AI can describe an old version of your pricing, features or even your name.
The general method is in how to check if AI recommends your business. This guide adapts it for a product.
Memory versus live search matters more for new products
A model's memory stops at its training cut-off. Anthropic, for example, lists a June 2026 knowledge cut-off for its current Claude models (Anthropic). A product that launched or changed after a model's cut-off cannot be in its memory. It can only be found if live search reads a current page.
So test both ways: with search on (can it find you today?) and with search off (does it remember you?). The difference is explained in memory versus live search.
Which index each AI reads when search is on
| AI | Live search draws on | How sure we are |
|---|---|---|
| ChatGPT | Several providers; OpenAI has not said how much comes from Bing | Partly confirmed (OpenAI) |
| Claude | Brave Search, per Anthropic's subprocessor list | Reported, not confirmed in documentation (TechCrunch) |
| Gemini | Google Search's index | Confirmed (Google) |
| Perplexity | Its own index | Confirmed (Perplexity) |
The full evidence is in which search engine each AI uses.
How to check, step by step
- Write three sets of questions. - Problem questions, no brand: "best project management software for a design studio", "what tool should a small clinic use for appointment reminders". - Comparison questions: "alternatives to [a competitor]", "[a competitor] vs [another competitor]". - Questions that name you: "what is [your product]?", "how much does [your product] cost?", "what does [your product] integrate with?"
- Use a private window. A clean session keeps your history from colouring the answer.
- Turn web search on in each AI and look for source links in the answer.
- Ask each question three times on each AI. Keep every answer.
- Record for each run: named (yes or no), website linked (yes or no), how the product is described, competitors named, and sources listed.
- Ask again with search off. That is the memory check.
What the sources tell you
The most useful column in your record is the sources. For software questions they are often a mix: vendor sites, comparison pages, review sites, forum threads, articles. In one informal test of three buying questions in ChatGPT, run by a marketing agency that sells Reddit services, Reddit appeared in all three answers while G2, Capterra and comparison pages each appeared once (Foundation Marketing). That is illustrative, not a study, and the same piece reports Reddit's share of ChatGPT citations fell sharply in August 2026. The lesson is to look at your own questions, not at a general rule.
For each competitor that is named, open the pages listed. Those pages are where the AI found them, and so are the places you need to be present.
Check the description, not only the mention
Read each sentence about your product against the truth: what it does, who it is for, price, integrations, and whether it still exists under that name. A wrong sentence usually comes from a stale page. Fix the page you control, and ask the owner of any third-party page to correct it.
Check the crawlers
Open yourproduct.com/robots.txt and confirm that the AI search crawlers are not blocked by mistake. OpenAI says a site that opts out of OAI-SearchBot will not appear in ChatGPT search answers (OpenAI). Perplexity says to allow PerplexityBot to appear in results (Perplexity). Anthropic lists Claude-SearchBot for search (Anthropic). Also make sure your pages render their text without needing JavaScript to show it, since a plain page is easier for any crawler to read.
Answers move, so check on a schedule
The same question can return different products from run to run. Count how many of three runs name you, and repeat after each launch. See why AI answers change every time.
Or have it done every morning
aipozi asks the questions you choose on ChatGPT, Claude, Gemini and Perplexity every morning with live search on, and records whether your product was named, whether your site was linked, and which competitors were recommended instead.
Frequently asked questions
Can I check if AI recommends my software product for free? Yes. ChatGPT, Claude, Gemini and Perplexity all have free versions. Ask each the questions a buyer would type, with web search on, and record whether your product is named. Free plans may cap searches, so spread the four AIs across separate sittings.
Why does an AI describe my product as something it is not? It is working from pages it read or from memory, and either may be out of date. Models also have a training cut-off, so a product that launched or changed after it can only be described correctly if search finds a current page. Open the sources the AI listed to see what it read.
Do review sites and Reddit matter for AI recommendations? They can appear among the sources AIs cite for software questions, but studies disagree on how much, and the vendors running them often sell related services. Check the sources AIs list for your own questions rather than relying on a general figure.
Should I include my competitors' names in the questions? Yes, as a separate set. Questions such as "alternatives to [competitor]" show whether you are named when a buyer is already comparing. Questions with no brand names show whether you are named by someone who does not know any product yet.
How often should I check? Monthly, and after a launch, a rename or a pricing change. Answers change between runs and model updates, so one test is only a snapshot. For what to change when you are missing, see how to get recommended by ChatGPT.