How to check if ChatGPT recommends your product

"Does ChatGPT recommend us?" sounds like a yes-or-no question. It isn't. Ask ChatGPT the same buyer question twice, or phrase it slightly differently, and the shortlist can change. This guide shows how to check properly, by hand or with a tool, so you get an answer you can act on instead of a screenshot that proves nothing.

Why one chat isn't an answer

Our own data shows how much answers move. In our AI rankings run on 26 October 2026, we asked a ChatGPT model with web search 5 buyer questions in each of 20 SaaS categories:

• Across the 20 categories we recorded 264 product-and-category pairs. 151 of them (57%) appeared in only one of the five answers for their category.
• Only 9 products were named in all five answers for their category: Mixpanel, Amplitude and Heap (product analytics), Bitwarden, Dashlane and Keeper (team password managers), Buffer (social media management), and UptimeRobot and Better Stack (uptime monitoring).
• 83 of the 100 answers cited web sources; 17 named products without citing anything.

So if you ask once and see your product, that doesn't mean ChatGPT reliably recommends you. If you ask once and don't, that doesn't mean it never does. A useful check uses several questions, records the same things every time, and repeats.

Step 1: Write the questions your buyers actually ask

Start with 10 to 25 questions in your buyers' words, not your marketing words. Cover these patterns:

1. Best-of: "What's the best [category] for [audience]?"
2. Shortlist: "I'm choosing [category] for [audience]. Which products should I shortlist and why?"
3. Popularity: "What [category] do most small teams use right now?"
4. Constraint: "Recommend an affordable [category] with good support." Or swap in your real differentiator: an integration, a region, a team size.
5. Ease: "Which [category] is easiest to set up for a small team?"
6. Alternatives: "What are the best alternatives to [competitor]?"
7. Comparison: "[You] vs [competitor]: which is better for [use case]?"

Constraint questions matter more than you'd think. In our data, the "affordable … with good support" question put a product outside the category's top 3 in first place in 16 of 20 categories, for example Hesk for help desks, Patriot Accounting for small-business accounting, and Less Annoying CRM for startup CRMs. If you're a smaller product, these narrow questions are often where you'll show up first.

Keep the wording fixed once you've chosen it. Changing the questions every time makes trends impossible to read.

Step 2: Ask in a clean session

ChatGPT can personalize answers, and your history shouldn't skew a check. As general practice:

• Use a Temporary Chat or a logged-out session, so saved memories and past conversations don't steer the answer.
• Don't mention your product or competitors earlier in the same conversation. Ask each question in a fresh chat.
• Make sure web search is used if you want to see sources. Answers with search are the ones that cite pages you can influence. If you want both views, ask with and without search and record them separately.
• Note the date and, if visible, the model. Behavior changes between versions.

Step 3: Record the same five things every time

For each answer, write down:

1. Named or not. Is your product mentioned as a recommendation?
2. Position. If named, is it first, in the middle, or last? Count the order of distinct products.
3. Competitors named. Every product in the answer, in order.
4. Sources cited. Every domain linked in the answer.
5. How you're described. Is the description accurate? Wrong pricing or an old name is a fixable problem.

Position deserves its own column. In our email marketing data, Brevo was named in 4 of 5 answers but never first, while Mailchimp was named in 3 and came first in 2. In live chat, Crisp and Tidio were each named in 4 of 5 answers but never first, while Tawk.to was named in 3 and came first in all 3. Frequency and position tell different stories.

Step 4: Turn answers into three numbers

Once you have a batch of answers, summarize them:

• Share of answers: the answers that name you, divided by the questions asked. "Named in 7 of 20" is a baseline you can track.
• Average position when named: lower is better. Being named first in a few answers can matter more than being an afterthought in many.
• Top cited sources: the domains that appear most often across answers, especially in answers that name competitors but not you. That list is your outreach and content plan.

Then put your competitors through the same numbers. Your share means more next to theirs.

Step 5: Read the sources, not just the names

The cited sources tell you *why* the answer looks the way it does. In our 26 October run:

• The 100 answers linked 424 different websites (650 links in total, counting image and link cards). Only 204 of those sites were cited inline to back a specific claim; the rest appeared only as cards.
• techradar.com appeared in 32 answers across 13 categories, the most of any site. toolradar.com was next, in 14 answers across 9 categories.
• G2 appeared in 3 answers. Reddit and Capterra didn't appear in any answer in that run.
• Some links were odd: the logo-image site freepng.com appeared in 8 answers, but only as an image card, and never backed a claim. Read what a citation actually is before you treat it as a signal.

If a handful of editorial lists keep appearing for your category and you're not on them, you've found the main reason you're missing. Our guides on ChatGPT and why ChatGPT recommends competitors cover what to do next.

Step 6: Repeat on a schedule

A single batch is a snapshot. To see whether anything is working:

• Re-run the same questions weekly or every two weeks, in clean sessions.
• Keep a simple log of what you changed (a new comparison page, a listicle placement, an updated directory profile) with dates.
• Look at trends over four to eight weeks, not week-to-week jumps. Small samples are noisy.

Doing it by hand vs. using a tool

Checking by hand works and costs nothing but time. For 20 questions it takes an hour or two each round, and the hard part is consistency: the same wording, clean sessions, and careful recording.

A tool does the same thing automatically. ShortlistWatch's free check runs 5 buyer questions for your category through ChatGPT with web search, and shows:

• whether you're named, and your position;
• which competitors get named;
• which sources the answers cite.

It needs no account. Paid plans re-ask 25 questions per product every week and track share of voice, sources and trends over time. Whichever way you do it, the method is the same: fixed questions, clean sessions, the same records, repeated.

Common mistakes

• Asking once and screenshotting. It proves very little either way.
• Asking in your own long-running chat. Your history and memory can shape the answer.
• Only asking "best [category]". Broad questions favor incumbents. Add constraint, audience and alternatives questions.
• Ignoring position. "Mentioned last" and "recommended first" are not the same result.
• Ignoring sources. They're the part you can change.

The short version

1. Write 10 to 25 fixed buyer questions, including narrow constraint questions.
2. Ask each in a clean session, with web search, and note the date.
3. Record named or not, position, competitors, sources and description.
4. Summarize share of answers, average position and top sources, for you and your competitors.
5. Use the sources to plan outreach and content.
6. Repeat on a schedule and read the trend, not the single result.