A practical 5-minute method to see if ChatGPT recommends your business — discovery vs branded prompts, a results worksheet, and when to use a free multi-engine report.
Here's the short answer: open ChatGPT in a private window, ask the same question a customer would ask without naming your brand, include your city or niche, and write down whether you were named, recommended, where you sat in the list, who else showed up, and whether the facts were right. Do that for five to ten prompts. It takes about five minutes and tells you more than a single branded vanity search.
If you want the same check across ChatGPT, Perplexity, Gemini, Copilot, and Google AI Overviews without copy-pasting prompts, Sift runs a free multi-engine report and emails you the results. This guide is the manual method — useful either way, and free.
Two different questions (don't mix them up)
Most people type “Does ChatGPT know [my brand]?” That only tests whether the model has heard of you. Customers almost never ask that. They ask for a recommendation.
- Discovery question: “Best [category] in [city/region] for [need]” — this is the one that wins or loses customers.
- Branded knowledge question: “What is [your brand]?” or “Tell me about [your brand]” — this checks whether the assistant can describe you accurately once someone already knows your name.
You can be famous to the model and still invisible on discovery prompts. Run both. Treat discovery as the score that matters for new demand.
The 5-minute manual ChatGPT check
- Use a clean context. Open an Incognito/Private window (or log out). Old chats and personalization can nudge answers toward brands you've mentioned before.
- Turn search on if the product offers it. ChatGPT with web browsing reflects what it can retrieve right now, not only training memory. That's closer to what a customer sees.
- Ask like a customer. No brand name. Add location, use-case, budget band, or industry if that's how people shop for you.
- Ask again with a small twist. Change the city, the adjective (“affordable” vs “premium”), or the outcome. One answer is a snapshot; lists move between runs.
- Log the result in a spreadsheet. Named? Recommended? Position? Who else? Which sources? Accurate? Date? (Worksheet below.)
- Only then run a branded prompt to see what ChatGPT says when someone already knows your name.
Named is not the same as recommended. Being listed third with a shrug is different from being the clear pick with reasons.
Ready-to-copy prompt templates
Swap the brackets. Keep the structure. Ask five to ten variants, not one.
| Audience | Discovery prompt | Branded follow-up |
|---|---|---|
| Local service | “Who are the best [plumbers / dentists / lawyers] in [city] for [specific need]? Give 5 options with reasons.” | “What do you know about [Brand] in [city]? Is it a good fit for [need]?” |
| Local restaurant / retail | “Recommend [cuisine / product] places in [neighborhood] for [occasion]. Rank them.” | “Summarize reviews and what [Brand] is known for.” |
| B2B software | “Best [category] tools for [company size / industry] in 2026. Compare free vs paid.” | “What is [Brand]? Who is it for, and what are the limits?” |
| Professional services | “Which [agencies / consultants] in [region] are good at [outcome]? Include smaller firms.” | “Describe [Brand]'s services and typical clients.” |
| Ecommerce / DTC | “Best [product type] for [use case / skin type / budget]. Which brands should I shortlist?” | “Is [Brand] reputable? What do buyers usually like or dislike?” |
Tip: if your market is small, add the country or language. If it's crowded, add a constraint customers actually use (“same-week appointment,” “SOC 2,” “under €50”).
Results worksheet (copy into Sheets)
One row per prompt per assistant. After a week you'll see patterns instead of vibes.
| Question | Assistant | Named? | Recommended? | Position | Also named | Sources | Accurate? | Date |
|---|---|---|---|---|---|---|---|---|
| Best … in … | ChatGPT | Y/N | Y/N / soft | 1–n / none | competitors | URLs cited | Y/N + notes | YYYY-MM-DD |
Named vs recommended vs cited
- Named: your brand appears in the answer at all.
- Recommended: the assistant presents you as a good fit, with reasons — not just a name in a laundry list.
- Cited: a specific URL (yours or someone else's) is shown as a source. Citations show what the answer is leaning on.
Aim for recommended + accurate. Named-but-wrong (old address, dead product, competitor attributed to you) is worse than silence — fix the underlying pages and third-party listings the assistant is reading.
Why the answer changes when you re-run it
Treat every answer as a snapshot. Assistants pull from different source mixes, re-rank between runs, and personalize when you're logged in. Industry research used in our own playbooks keeps finding the same pattern: each assistant leans on a different slice of the web, and one good ChatGPT answer says little about Google AI Mode or Perplexity.
So: log three runs on different days before you decide you're “in” or “out.” If you're never named across those runs, you have a real gap — not bad luck on one Tuesday morning.
Don't stop at ChatGPT
Customers bounce between assistants. A fair check covers the ones your buyers actually use.
| Assistant | Good for checking | Tip |
|---|---|---|
| ChatGPT | Broad consumer + SMB recommendations | Private window; enable search/browse |
| Perplexity | Source-heavy answers with links | Note which domains it cites |
| Gemini | Google-ecosystem visibility | Compare with AI Overviews on the same query |
| Copilot | Bing-influenced answers | Useful if your buyers live in Microsoft 365 |
| Google AI Overviews / AI Mode | Search-adjacent recommendations | Same query in Google; watch the overview box |
Doing this by hand across five engines gets tedious fast. That's the job of an AI visibility tool. Sift's free report runs the multi-engine pass for you and returns a plain-language action plan.
What our free reports have shown so far
From an anonymized sample of 21 historical Sift reports (businesses that already asked for an AI-visibility check): visibility scores sat between 45 and 52 (mean about 48.7, median 49). Fourteen of the twenty-one scored under 50. Nineteen of those reports included ChatGPT among the engines checked.
That's a small sample, not a market study — but the pattern is useful: even among people motivated enough to run a check, mid-40s to low-50s scores are common. A weak or missing ChatGPT recommendation is ordinary, not a freak outcome. The point of the worksheet above is to see your gap clearly before you spend a month guessing.
If you're missing: what to do next (high level)
A full “how to get recommended” playbook is a separate article. The short version, grounded in the same research we use in Sift reports (and detailed in our GEO vs SEO guide):
- Make sure crawlers can reach you (robots.txt, CDN/firewall, indexation in Google and Bing).
- Show up on the third-party sites assistants already cite in your category — directories, review sites, partner pages — with consistent name and facts.
- Collect real reviews where your buyers already look.
- Publish answer pages on your own site (pricing, specs, comparisons) with the answer first, in real HTML text.
- Re-run this check in 2–4 weeks. Visibility moves, but not overnight.
If you want a structured baseline across engines today, start at usesift.net— free, no card, report by email.
FAQ
How do I check if ChatGPT recommends my business?
Use a private ChatGPT session, ask customer-style discovery questions without your brand name, include location or niche, and log whether you were named, recommended, and ranked. Repeat a few times on different days.
Does ChatGPT recommend businesses, or only describe them?
Both. Discovery prompts (“best X in Y”) produce ranked or shortlisted recommendations. Branded prompts mostly test whether the model can describe you. Measure both; optimize for discovery.
Why am I not showing up in ChatGPT?
Usually the assistants don't see enough clear, consistent mentions of you on the sources they trust for that category — or your key pages aren't reachable/indexable. One missing run isn't proof; never appearing across several clean runs is.
Is a branded search enough?
No. Customers looking for a provider rarely start with your name. If you only test branded prompts, you'll overestimate how often you win new demand.
How often should I re-check?
For a manual log, weekly is plenty while you're actively improving; monthly once you have a baseline. Re-check after major site, review, or PR changes.
Can I check Perplexity, Gemini, and Google AI Overviews the same way?
Yes — same discovery prompts, same worksheet columns. Expect different shortlists and different citations; that's normal.
What's the fastest way to check multiple AI assistants at once?
Use a multi-engine checker. Sift sends a free report covering the major assistants so you're not copy-pasting prompts all afternoon.