In one afternoon, AI assistants recommended Zapier 16 times and linked to it 0 times.
We ran the same test on Mailchimp. Then on Trello. Same story both times.
The bots named all three brands, over and over, and pointed a grand total of nobody at their websites.
That should stop you cold. Because every SEO blog on the internet is telling you the same thing right now: getting cited by an AI, which means named with a link back to your site, is the new ranking first on Google.
So we tested it. Real brands, real buyer questions, one afternoon. Here's what actually came back.
Count your links, not your mentions
Here's how the test worked.
We used our own AI Visibility Checker. You give it a website. It works out what the company sells, then writes 6 questions a real buyer would type before they've heard of you. Things like "what's the best way to connect my apps without code?"
It sends those 6 questions to 6 different AI assistants. And it counts two things, on purpose kept apart: how many answers named the brand, and how many actually linked to it.
One rule does the heavy lifting. It won't let your brand name into the question. Ask "is Mailchimp any good?" and of course the AI says Mailchimp straight back. You've measured your own typing, nothing more.
Here's what came back.
| Brand | Answers that named it | Platforms it showed on | Links back to it |
|---|---|---|---|
| Zapier | 16 of 24 | 4 of 6 | 0 |
| Mailchimp | 11 of 30 | 5 of 6 | 0 |
| Trello | 6 of 24 | 2 of 6 | 0 |
Run your eye down the "named" column. Now down the "links" column.
The first one jumps around. The second one is a wall of zeros.
78 answers. 33 recommendations. 0 links.
And that gap is the whole problem, sitting right there in two columns.
A mention feels like a win. An AI just told a stranger that your brand is the answer. But walk it forward and ask what actually reaches you.
Nothing does.
No click. No visit. No line in your analytics, the report that counts who showed up. You can't trace a single signup back to it. It's a billboard on a motorway: seen by thousands, pulled into by none.
A link is the opposite. Someone can click it and land on your site. That's the thing search has quietly sold you for twenty years.
Here, it happened zero times out of 78.
So keep the two numbers apart, and give them to two different people. Mentions are the brand team's problem. Links are the traffic team's problem.
Blend them into one cheerful "AI visibility" score and you've buried the only half that pays. Then six months later someone asks why the number keeps climbing while the traffic stays flat, and you get to work that out live, in the room.
Stop expecting your brand to save you
Look at the table one more time, but by size this time.
Zapier is the giant. Trello is the small one. Mailchimp sits in between. And the mentions line up exactly like that: big brand named most, small brand named least.
So the plan writes itself, right? Get bigger. Get famous enough and the robots can't help but quote you.
Then look at the links again.
Zapier pulled almost 3 times the mentions of the smallest brand here. It earned the exact same number of links. Zero. Ten years as a household name in its category bought it precisely nothing in that column.
That quietly kills a plan a lot of teams are running today.
Whatever makes an AI link to you is simply not whatever makes it name you. Two different levers. And you cannot pull the second one by yanking harder on the first.
Find out whose pages it cites instead
Here's the line that changed how I read the whole test.
Every single report ended the same way:
"None of your pages were cited in these answers. The AI is answering this category from other sites instead."
Read that twice.
The bots weren't stuck. They weren't short of information. They answered all 6 questions, confidently, every time. They just took the answer from somewhere that wasn't the brand's own website.
So the real question in AI search isn't "does the AI know me?" It plainly does. It named these brands 33 times.
The question is this: whose pages is it reading when someone asks about the thing you sell?
And unlike a fuzzy visibility score, that one has an answer you can go and read tonight.
Open ChatGPT. Ask it a question about your market, with your name left out. Then look hard at every source it links to.
Some will be review sites. Some will be your competitors. Some will be a random forum thread from 2019. Whatever they turn out to be, those are the pages standing between you and your next customer.
That list beats any AI visibility score you'll ever be sold. And it costs you ten minutes.
Getting an AI to say your name is easy. Zapier managed it 16 times. Getting the link, the part that actually sends you a visitor, is the hard bit. That's the whole job of GEO (getting your pages quoted, with a link, inside AI answers), and it's what we do.
- We measure who AI names you to, and who it links you to
- We rebuild the pages AI can't quote yet
- SEO and GEO run as one plan, not two
- A free call to see exactly where your gap is
Check you're not blocking the bots yourself
Two weeks after this test, we ran a second one. It turned up something worth bolting straight onto this article.
We opened the robots.txt of 65 big-name companies. It's the file that lists which pages you let bots read. The full study is right here.
Quick word first, because it matters. A crawler is the bot an AI sends out to read your pages so it can learn what you do.
Here's the split we found. Every news publisher we checked blocks those crawlers. Not one software company does.
The publishers have a fair point. If a machine can answer in your place, why hand it your work for free?
But flip that around for a second. If people have to find you before they can buy from you, blocking the crawler doesn't protect you. It deletes you from the answer. And the answer already wasn't linking you.
There's a sharper twist underneath it. The one crawler that fewest companies block is the exact one that hands readers a link back.
So before you decide the AIs are snubbing you, go and open your own robots.txt.
Sometimes they ignore you because you told them to, in a file somebody edited two years ago for a reason nobody at the company can remember.
Run this two-number check every quarter
Here's the whole job. It takes an afternoon, and you can start it today.
Run the check. Write down your two numbers: how many answers named you, how many linked you. Do it again next quarter. Put the two side by side.
If the mentions climb and the links stay stuck at zero, your brand team is winning and your traffic team isn't. That's a PR result. Report it as one, and stop filing it under SEO.
If a link finally shows up, drop everything and study the page that earned it. That page is your template. It'll teach you more than any AI-search guide, this one included.
And don't let the real work slide while you watch the robots. Google is still where the clicks live.
Our free Organic Traffic Checker, which shows how many visitors a site pulls from search, gives you today's picture. What it can't show is the two-year trend, and that's usually the number your boss actually wants. Semrush and SE Ranking both keep that history.
One last thing worth reading, because it's a strange mirror of this one. We dug into where Zapier's search traffic really comes from, and almost all of it traces back to a single blog post. Not the 140K programmatic pages everyone points at.
Two numbers, counted apart, from 6 buyer questions with your brand name deliberately left out. Most teams have never checked either one.
- The same check that found 33 mentions and 0 links
- Questions with your brand name in them are refused
- 3 checks a day, no signup, no card
What we could not measure
Here's what this test doesn't show. Some of the tools we name are partners of ours. If you buy through those links we earn a commission, and you don't pay a cent more.
This is 3 brands on one day. It's not a study. We asked 6 questions per brand and collected 78 answers.
That's enough to prove these three got named constantly and linked never. It isn't enough to hand you a rate for any market in general. If you want a number for your own site, run it on your own site. That's the whole point.
We measured on 23 August 2026 and haven't re-run it. Assistants change how they link without telling anyone, so treat that zero as a snapshot from one day, not a law of nature. The robots.txt numbers are from a couple of weeks later.
We only count a link when it points at the brand's own website. If an AI named a brand and linked a review site instead, we logged that as a mention with no link. That's exactly the case we're describing here. So "0 links" means "0 links to them", not "0 links at all".
And two honest faults in our own tools, straight from us. The AI Visibility Checker works out your category from your site, and now and then it invents one nobody would type. It once produced "Salaries Tools" for a company that sells nothing of the sort. On some runs, 3 of the 6 assistants come back as "not reached", so a low platform count can be our failure rather than your absence.
Frequently asked questions
Do AI assistants link to the brands they recommend?
Not in this test. We asked 6 buyer questions about 3 products and collected 78 answers on 23 August 2026. The brands got named 33 times and linked 0 times. Every report ended with the same line: none of your pages were cited in these answers.
What's the difference between a mention and a link?
A mention is the AI saying your name. A link is something a reader can click to reach you. Only the link sends a visitor, and only the link shows up in your reports. Rolling both into one "AI visibility" score hides the half you can act on.
Which brands did you test?
Zapier, Mailchimp and Trello. Zapier got named most often, in 16 of its 24 answers. Trello least, in 6 of 24. All three got zero links back to their own websites.
Does a bigger brand get linked more?
No. It gets mentioned more and linked the same. Zapier got named nearly 3 times as often as Trello, and both got 0 links. So whatever earns a link isn't what earns a mention, and growing your brand won't close the gap.
Why can't the question include my brand name?
Because handing a model your name guarantees it says the name back, which measures nothing. Our checker refuses those questions and asks what a buyer would ask before they've heard of you. If a tool reports a high score, make it show you the exact questions it used.
Should I block AI crawlers?
It depends on whether a machine replaces your product or advertises it. We read the robots.txt file of 65 companies in a separate study. Every news publisher blocked AI crawlers, and no software company did. The one fewest sites block is the bot that hands readers a link.
Where do AI answers get their information?
In this test, from sites other than the brands being named. The models answered all 6 questions confidently and just sourced them elsewhere. So the useful question isn't "does the AI know me", it's "whose pages is it reading about my market".
How often should I run this?
Once a quarter is plenty. Write down both numbers each time. Assistants change how they link without any warning, so any single reading, including this one, is a snapshot rather than a rule.
The one number to take away
0.
An AI told somebody that one of these 3 brands was the answer 33 separate times. It gave them a way to actually get there 0 times.
The whole industry has spent 2 years bracing for a world where being quoted by AI replaces being ranked by Google. In this test, the quoting-with-a-link half didn't happen. Not rarely. Not at a trickle. Not once. Meanwhile the naming half happened non-stop.
Both numbers are real, and only one of them is traffic. If you're going to report AI visibility to anyone, report the two apart. A chart that adds them together tells your team something's working, right up until the day someone checks the part that pays.
Measured 23 August 2026 with our own AI Visibility Checker: 6 buyer questions per brand, none containing the brand name, sent to ChatGPT, ChatGPT Search, Claude, Gemini, Gemini Search and Perplexity, with mentions and links counted apart. The robots.txt comparison is from 6 September 2026, and its method is in our repo. The tools are free, so you can run the same check on your own brand.