By KeywordSpy Team · Last updated: July 2026 · 13 min read
In This Guide
What a good result looks like: the 0-100 AI Visibility Score
Why your brand may be invisible in your own language or market
To check whether AI recommends your brand, you ask the AI engines the category questions your customers ask - without naming yourself - many times over, and measure how often you appear. Do that once and you learn almost nothing: one SparkToro and Gumshoe.ai study found less than a 1-in-100 chance of getting the same AI brand-recommendation list twice, so a single check is closer to a coin flip than a verdict.
This guide gives you the full free method to run the check yourself across ChatGPT, Gemini and Perplexity, shows you how to read the result honestly, and is written for the in-house marketer who needs one number to report, the agency proving a client gap, and the non-English founder who suspects the English tools are undercounting them.
Quick Facts
| Question | Short answer |
|---|---|
| How do you check? | Ask category buying questions with no brand name; measure how often you appear |
| Which engines to test | ChatGPT, Gemini, Perplexity (add Claude and Mistral if you can) |
| How many times to repeat | 5-10 runs per prompt, per engine (answers are non-deterministic) |
| Same list twice? | Less than a 1-in-100 chance (SparkToro / Gumshoe.ai, 2025) |
| Scale of the channel | ChatGPT reached ~800M weekly users by Oct 2025 (TechCrunch) |
| Is it Google SEO? | No - a separate channel, and separate again from Google AI Overviews |
| Cost to check | Free by hand; free first scan with KeywordSpy, no account |
What it means for AI to recommend your brand
Ask an AI assistant "what's the best tool for X" and you don't get ten blue links. You get a short list of a few names, sometimes with a sentence on each. If your brand isn't in that list, you're not in the conversation - and the buyer often stops there.
That is what AI visibility means: whether the models name you when someone asks a buying question in your category, how you stack up against the competitors named alongside you, and where you land in the order. It breaks down into three plain parts. Presence is whether you appear at all. Share of voice is how often you appear compared with rivals across many runs. Position is whether you're the first name or the fifth. A brand can score well on one and badly on another, which is exactly why a yes-or-no answer misses the point.
This channel is not a niche. ChatGPT alone reached 800 million weekly active users by October 2025 and was reported near 900 million by early 2026, and Gartner expects traditional search engine volume to fall as answer engines take over more of the query. When the answer replaces the search results page, being named in the answer is the whole game.
So "does AI recommend your brand" is a real, measurable question with a real business cost attached. The rest of this guide is how you get the number.
How to check whether AI recommends your brand, step by step
You can do this today, for free, in about an hour, and you don't need a tool to start. The manual method below gives you an honest first read on every major engine, and it teaches you two things a dashboard won't: exactly which questions your buyers ask, and which rivals the models reach for when your name doesn't come up. Here's how to run it.
Write your buying prompts, not your brand. List 20 to 30 questions a customer asks when they're close to choosing, phrased the way a real person types them. Think "best CRM for a small agency" or "affordable project management tool for remote teams." Do not mention your own name - you want the questions where the AI decides who to recommend.
Open a clean session in each engine. Use ChatGPT, Gemini and Perplexity at a minimum. Log out or use a private window so your chat history and account personalization don't feed you a flattering answer that no one else sees.
Run each prompt and record the raw answer. Note three things every time: did your brand appear (presence), which other brands were named (your real competitor set), and where you sat in the order (position).
Repeat, then repeat again. Run each prompt five to ten times per engine. This is the step almost everyone skips, and it's the one that matters most, for reasons the next section makes clear.
Count your hit rate. For each engine, divide the number of answers that named you by the total runs. If you appeared in 3 of 10 ChatGPT answers, that's a 30% presence rate on that engine. Do the same for Gemini and Perplexity.
What you end up with is a rough visibility rate per engine, plus a list of the competitors the models keep naming instead of you.
That competitor list is often the most useful part. It tells you who owns the answer today, and it's frequently not the brands you'd have guessed.
If you'd rather not run 200-plus prompts by hand, you can run a free scan on KeywordSpy for one real check with no account and results in about 15 seconds. Either way, the method is the same. The tool just does the repetition for you.
Why checking once gives you the wrong answer
Here's the trap. You ask ChatGPT once, you're named, and you conclude you're winning. Or you're absent once and you panic. Both readings are wrong, because AI answers are non-deterministic - the same prompt returns different lists on different runs.
The scale of that swing is bigger than most people expect. The SparkToro and Gumshoe.ai research ran roughly 2,961 prompt runs across ChatGPT, Claude and Google AI with 600 volunteers and found less than a 1-in-100 chance of getting the identical brand list twice. Penn State researchers pushed it further: across 10 identical runs with maximum-consistency settings, one model swung from 88% to 44% presence, and another from 75% down to 3%. Same question, same settings, wildly different answers.
That's why a single check is noise, not signal. Your real visibility is a rate, not a yes or no - and you only see the rate by running each prompt many times and averaging, or by tracking it continuously so you catch the trend instead of one lucky or unlucky roll.
What a good result looks like: the 0-100 AI Visibility Score
Once you have a hit rate, you need to know what's good. There's no single pass mark, because a crowded category behaves differently from a narrow one, but presence is always the first hurdle. If you're named in most answers for your core buying prompts, you're in a strong spot. A few, and you have a foothold worth defending. None, and you're invisible in the channel your buyers are starting to use.
KeywordSpy rolls those three parts - presence, share of voice and position - into a single AI Visibility Score from 0 to 100, and shows all three so it's not a black box. You can build the same read by hand from your manual counts. Here's a simple way to interpret where you land.
| Score band | What it means | Typical situation |
|---|---|---|
| 0-10 | Effectively invisible | Named in almost no answers; competitors own the category |
| 11-30 | Occasional mention | You surface sometimes, usually low in the list, rarely first |
| 31-60 | Real foothold | Named in a fair share of answers; mid-pack share of voice |
| 61-85 | Strong presence | Named in most answers, often near the top of the list |
| 86-100 | Category leader | Named almost every time, frequently the first brand cited |
Read the score as a starting line, not a grade. What you want is the trend: the same prompts, checked the same way, moving up over weeks. A number you can't compare against last week's number can't tell you whether anything you did worked. If you want your own score to benchmark against, the free scan returns one in about 15 seconds, and you can see the whole score history on a weekly tracking plan.
How this is different from Google SEO and AI Overviews
This is where a lot of advice gets muddy, so let's be precise. There are three different things here, and they are not the same channel.
Google SEO is about ranking pages in Google's blue-link results. AI visibility is about being named inside the answers that ChatGPT, Gemini, Perplexity, Claude and Mistral generate. And Google AI Overviews are a third thing again - the AI summaries that sit on top of Google Search, tied to Google's own index. You can rank on page one of Google and still be absent from every AI engine, because the models are reading and weighting sources differently than Google ranks them.
Same brand, three different scoreboards.
Why it matters now: Gartner predicts traditional search engine volume will drop 25% by 2026 as AI chatbots absorb queries that used to go to a results page. If you only track Google rankings, you're measuring a channel that's shrinking while ignoring the one that's growing. Track the AI engines as their own thing, with their own score, separate from your SEO reports and separate from AI Overviews.
Why your brand may be invisible in your own language or market
Most AI visibility checks are run in English, and that quietly breaks the result for everyone who doesn't sell in English. If your customers ask their buying questions in Latvian, Lithuanian or Estonian, an English-only check is testing the wrong prompts in the wrong market.
Three things go wrong the moment you leave English. Your brand name takes grammatical (inflected) forms, so a plain string match misses the mentions where the name is bent by the local language. Your competitor set is local, not the global list an English query surfaces. And buying intent carries local geography that an English prompt strips out.
Miss those and you'll read yourself as invisible when you're actually being named, or the reverse.
This is the gap KeywordSpy was built to close. It's local-native rather than an English tool with translation bolted on: it matches inflected forms of your brand name, knows the local competitor set for the market you're tracking, and queries in your customer's own language with local geographic intent, across EN, LV, LT and ET. If you sell outside English, check in the language your buyers actually use - that's the only measurement that reflects your real market.
Manual checks vs tracking tools
The manual method is genuinely free and worth doing at least once - it teaches you your real competitor set and the prompts that matter. The catch is repetition. Because answers swing so much between runs, a trustworthy read means hundreds of prompts, run the same way, again next week and the week after. That's where doing it by hand falls apart, and where a tool earns its place. Here's an honest comparison of the options.
| Option | Cost | Repeats and tracking | Local-language handling | Best for |
|---|---|---|---|---|
| Manual / DIY spreadsheet | Free (your time) | Manual; hard to repeat identically | Only if you build it yourself | A one-time snapshot and learning your prompts |
| English-first SaaS tools | Mid to high monthly | Automated | Weak; misses inflected names and local competitors | English-market brands |
| Enterprise platforms | High, often sales-led | Automated, deep | Varies; built for large global teams | Large enterprises with budget and analysts |
| KeywordSpy | Free first scan; plans from €49/mo | Automated weekly re-checks and score history | Local-native (EN, LV, LT, ET) by design | In-house marketers, agencies, non-English SMBs |
KeywordSpy checks ChatGPT, Gemini and Perplexity on every plan, adds Claude on Growth and Mistral on Pro, and shows the three score components rather than hiding them. The first scan is free with no account and no card. From there, €3.99 unlocks a 10-day trial, and plans run €49 (Starter), €99 (Growth) and €249 (Pro, with whitelabel reports for agencies). If you're an agency proving a client gap, you can run the free scan on the client brand, then sign up to track it weekly and drop the score into your reporting. Pick the pieces you need from the pricing page.
What to do if AI doesn't recommend your brand
If you ran the check and came up thin, don't panic - a low score is a starting point, and this channel is young enough to move fast. The models name the brands they can find clear, credible, first-hand information about. So the work is making yourself the obvious answer in the sources these models read.
Start with content that has a point of view and real substance: unique first-hand experience, data others don't have, and clean structure the models can parse. The evidence backs this up - a Princeton, Georgia Tech and Allen Institute study found that adding cited statistics, quotations and credible source citations lifted a page's visibility in AI answers by roughly 30 to 40%. That's the same recipe this article follows.
Be worth citing, and the citations follow.
One thing to skip: don't waste time on llms.txt or other AI-specific files as a shortcut. Google's own generative-AI guidance says those files don't help. What helps is the harder, realer work - being genuinely worth citing.
And whatever you do, measure it. Set your baseline score now, make your changes, and re-check the same prompts on a schedule so you can see the line move. Without the before-and-after, you're guessing.
Run a free scan - one real check, no signup, results in about 15 seconds. See exactly where you stand across ChatGPT, Gemini and Perplexity, and who AI names instead of you, at keywordspy.ai. If you want the weekly trend and the full local-language tracking, the plans start at €49/mo, and new accounts get 20% off the first three months with code START20. You can read how the whole thing works on the KeywordSpy homepage.
Frequently Asked Questions
How do I check if AI recommends my brand?
Ask ChatGPT, Gemini and Perplexity the buying questions your customers ask in your category, without naming your brand. Use a fresh, logged-out session so past chats don't skew the answer. Note whether your name appears, who gets named instead, and where you land in the list. Repeat each prompt five to ten times per engine, then count how often you show up. That share is your real answer.
Can I just check once to know if ChatGPT recommends my brand?
No. One check tells you almost nothing because AI answers are non-deterministic. A SparkToro and Gumshoe.ai study found less than a 1-in-100 chance of getting the same brand list twice, and Penn State researchers watched one model swing from 88% to 44% across identical runs. You can be named on your first try and vanish on the next. Repeat each prompt many times, or track it continuously, before you trust the result.
Is checking AI visibility free, or do I need a paid tool?
Checking is free if you do it by hand. Open each engine, run your prompts, and log the results in a spreadsheet. That works for a one-time snapshot. The cost is time and consistency, since answers shift week to week and manual runs are hard to repeat the same way. A tool like KeywordSpy automates the repeats, tracks the trend, and gives you one score. The first KeywordSpy scan is free with no account.
How is AI visibility different from Google SEO and AI Overviews?
AI visibility is whether ChatGPT, Gemini, Perplexity, Claude and Mistral name your brand in their own answers. That is a separate channel from Google SEO, which is about ranking blue links, and separate again from Google AI Overviews, which sit on top of Google Search. You can rank well on Google and still be missing from every AI engine, so treat and measure them as three different things.
What is a good AI Visibility Score?
It depends on your category, but treat presence as the first hurdle. Appearing in most answers for your core buying prompts is a solid position; appearing in a few is a foothold; appearing in none means you are invisible in that channel. The KeywordSpy AI Visibility Score runs 0 to 100 and blends three things you can see: presence, share of voice against competitors, and your average position in the list.
Why isn't my brand showing up in AI answers?
Usually because the models have little clear, first-hand information tying your brand to the category, or competitors are cited more often across the sources these models read. Thin content, no point of view, and few credible mentions all hurt. Skip AI-specific files like llms.txt, which Google says do not help. What helps is unique first-hand content, a clear point of view, cited data, and clean structure.
Can I check whether AI recommends my brand in my own language?
Yes, and you should. Most tools check in English only, so they miss inflected forms of your brand name, your local competitor set, and queries with local geographic intent. If your customers ask in Latvian, Lithuanian or Estonian, check in that language. KeywordSpy is local-native and matches grammatical name forms, local competitors and local-language queries, so a non-English brand is not undercounted.
Sources and Data
Gartner - press release, search engine volume forecast, 2024. gartner.com
Aggarwal et al. (Princeton / Georgia Tech / Allen Institute) - GEO study, ACM SIGKDD, 2024. arxiv.org/abs/2311.09735
TechCrunch - ChatGPT weekly active users report, 2025. techcrunch.com
SparkToro and Gumshoe.ai - AI brand-recommendation consistency study, 2025. sparktoro.com
Penn State research (via Popsight) - AI answer non-determinism, 2025. popsight.ai