Is AI Visibility Worth It? What the Numbers Actually Say
Falling traffic does not mean falling revenue, because AI is filtering out the browsers and leaving the buyers. That only helps you if you are on the shortlist, and shortlist position is decided by where you rank in ordinary search: a page at position 1 gets cited 58% of the time against 14% at position 10. So AI visibility is a layer on top of your search foundation, not a substitute for it, and the free AI Visibility Calculator hands back a range instead of one confident number because nobody can honestly promise you the rest.
I wanted to build a calculator that told you whether AI visibility work was worth paying for. So I went looking for the numbers everyone quotes.
Almost every one of them was published by a company that sells AI visibility services.
That is not a scandal. Vendors publish data, and some of it is fine. But the figures did not agree with each other. One said traffic from AI tools converts 4.4 times better than regular search traffic. Another said 5 times. Another said 23 times. Same claim, and a spread of roughly fivefold between people all selling the same thing.
When the numbers used to sell something disagree that badly, the honest move is not to pick the most flattering one. It is to go find out who measured without a stake in the answer.
What the neutral research actually found
The most useful data here did not come from anyone selling the service. Pew Research Center recorded the real browsing behaviour of 900 US adults across 68,879 Google searches in March 2025, then looked at what happened when Google showed an AI summary at the top of the results.
An AI summary is the block of text Google writes for you above the links, so you get an answer without visiting anyone's website.
Here is what they found.
When an AI summary appears, the clicks roughly halve
Share of Google searches that led to a click, March 2025
The 1% is the number worth sitting with. The source links inside an AI answer, the ones an AI visibility pitch is implicitly promising you a share of, are almost never clicked.
Source: Pew Research Center, 68,879 Google searches by 900 US adults, March 2025. Published 22 July 2025.
Clicks roughly halve. And the source links inside the summary, the ones every AI visibility pitch is implicitly promising you a share of, get clicked 1% of the time.
That data is from March 2025 and was published in July 2025, so treat it as a snapshot rather than today's weather. The direction has not reversed since.
Your traffic can fall while your revenue rises
Losing clicks is not the same as losing money, and conflating the two is the most expensive mistake in this whole conversation. Kevin Indig presented a client case where organic traffic fell almost 50% year over year while conversions rose 20%.
That is not a paradox once you see the mechanism. AI now answers the low-intent questions directly on the results page. The person who used to click your site to find out what a bezel setting is now gets that answer without visiting anyone. Those visits were never going to become a sale.
What is left after that filtering is people who already asked their questions, already compared their options, and arrived having chosen you by name. They convert very well, and that is the actual point. AI did the qualifying work that your top-of-funnel content used to do badly.
So the honest version of the vendor claim is not that AI traffic is magic. It is that AI removes the browsers and leaves the buyers.
One caveat worth stating plainly: that traffic-down, conversions-up figure is a single client. The recording is linked in the sources if you want it in context. Treat it as a well-observed mechanism rather than a benchmark you should expect to hit.
But only if you are in the answer
The filtering above is good news for exactly one group: stores that survive the shortlist. Everyone else gets the same traffic loss with none of the conversion gain, because AI absorbed their browsers and handed their buyers to somebody else.
The shortlist is smaller and more fixed than most people assume. Indig's figure is that ChatGPT draws from a pool of roughly 30 brands per topic, and about two thirds of all citations go to that pool. So this is not a long tail where everyone eventually gets a turn. It is a room with a limited number of chairs, and the question is whether you are in it.
You get to choose which room you are standing in
Here is where I disagree with how that finding usually gets repeated, and I have my own test to point at.
The day before that webinar, I ran six buyer questions against four AI engines using my own jewelry store as the subject. Two of them are the whole lesson.
Asked "where can I buy one of a kind pastel gemstone jewelry," the engines answered with six marketplaces and large retailers: 1stdibs, Brilliant Earth, Catbird, Etsy, Kendra Scott and Local Eclectic.
Asked "who makes one of a kind aquamarine jewelry," the answers named exactly one marketplace, Etsy, and cited my store in three of the four engines.
Same store. Same day. The only thing that changed was the verb.
One word changed who I was competing against
Marketplaces and large retailers named in the answer, same store, same day
The buying verb summoned 1stdibs, Brilliant Earth, Catbird, Etsy, Kendra Scott and Local Eclectic. The making verb summoned one marketplace, and cited my own store in three of the four engines. There is not one shortlist. There is a shortlist per question, and the question decides which room you are standing in.
Source: Andrea Li, first-party test on andreali.com across four AI engines (Google AI Overview, Google AI Mode, Perplexity, Gemini), 12 August 2026. Single dated snapshot, one run per engine; recommendations vary run to run.
The verb picks your competition before the ranking does
So the pool of thirty is real, but there is not one pool. There is a pool per question, and the buying verb summons the marketplaces while the making verb summons the makers.
If you are an independent designer trying to win "where can I buy," you have walked into a room where Brilliant Earth is already sitting down. The room where somebody asks who makes this is a room you can actually win.
This is the part I would push back on if somebody quoted the thirty-brand figure at you as a reason to give up. The number is not a verdict on whether you can compete. It is a description of one room, and you have some say in which room you are standing in, because the questions your pages answer decide which conversations you show up in.
That is a single dated snapshot, one run per engine, and recommendations vary run to run. I would run it on your own store before believing it applies to you.
That is the fork. Two jewelry designers can watch identical traffic declines in the same month, and one of them is being pre-qualified while the other is being replaced. The number in your analytics looks the same either way, which is what makes this so easy to misread.
So the question worth paying to answer is not how much traffic you can recover. It is whether you are in the answer when a buyer asks for someone who does what you do. That is what the calculator is asking you to put a value on, and it is why the value is a range rather than a promise.
AI visibility is a layer on top of SEO, not a replacement for it
Most designers get sold this backwards, and the correction is the tension underneath everything I publish. There is now hard data behind it.
A study of 16,851 queries and 353,799 pages, run by AirOps in partnership with Kevin Indig and published in April 2026, traced what decides whether ChatGPT cites a page. The finding was blunt. A page sitting at position 1 in the retrieval results was cited 58.4% of the time. A page at position 10 was cited 14.2% of the time. The report's own summary calls retrieval rank the number one signal, and describes it as a gap that no amount of content quality alone will close.
Where you rank decides whether you get cited
Citation rate by position in ChatGPT's retrieval results
This is why AI visibility is a layer on top of search rather than a replacement for it. The study calls retrieval rank the number one signal, and describes the gap as one that content quality alone will not close.
Source: AirOps in partnership with Kevin Indig, "The Fan-Out Effect," 16,851 queries and 353,799 pages, published 13 April 2026. Full report.
Read that again, because it is most of the argument. Where you rank largely determines whether you get cited. AI visibility work decides whether you are chosen from the shortlist. Your search foundation decides whether you are on the shortlist at all.
Trust is the one thing that beats rank
Rank is the dominant signal. It is not the only one, and the exception is the most hopeful finding here if you are a small studio competing against national retailers.
Indig's summary of his own research is that trust beats rank outright. You are not required to outrank a chain store to be cited ahead of one.
I want to be straight about why that finding landed hard for me. Twelve days before that webinar I wrote this, publicly, with no data in front of me: when everyone can build, being verifiably the real thing becomes the moat, which favours the maker over the clone, and the catch is that it only counts if the machines doing the recommending can actually find her.
I believed it because it is what I see at the bench and in client work. I did not have the measurement. Now somebody has run it across a hundred thousand citations and the measurement says the same thing, which is a better outcome than being right on instinct alone.
What trust actually looks like for a jeweler
For a jeweler, the trust signals are not abstract. They are your actual bench process shown rather than claimed, reviews from real buyers, a story about a specific piece that nobody could invent, and being described the same way everywhere the web mentions you.
That last one is duller than the others and does more work than any of them. If your studio is described one way on your site, another way on a directory, and a third way on an old marketplace profile, you are asking a machine to decide which version of you is real. Making those agree is unglamorous work with no immediate payoff, and it is the thing I most often find undone.
Here is the part I find genuinely encouraging, and I do not say that often about this subject. A catalogue business has to manufacture all of these signals. You have to publish what you already do. The bench photos exist. The process is real. The buyers are real people who wrote real things. Your disadvantage is scale, and scale is not what this rewards.
What you cannot do is skip both. A store with neither the ranking nor the trust signals is not in the running, and no amount of AI-specific tinkering changes that.
If you optimise for AI with no search foundation underneath, you are turning up an amplifier with nothing plugged into it. You can spend real money there and see almost nothing, then reasonably conclude the whole category is a scam.
It also explains why an honest calculator cannot promise you a number. The effect of the work depends on what is underneath it, and I do not know what is underneath yours. Any tool returning one confident figure is silently assuming a foundation you may not have.
The most useful thing you can do this week
The same study found that headings matching the buyer's actual question got cited 41% of the time, against 30% for headings that matched poorly. Heading structure mattered more than word count or how comprehensive the page was.
So write your headings as the question your buyer would ask. Not "Our Bespoke Process" but "How long does a custom engagement ring take?" Focused pages beat sprawling guides. That is a free change you can make today, and it works on both search and AI answers, which is exactly the kind of work I want you spending money on first.
Why I trust that study and not the multiples
Kevin Indig is an independent search researcher. He publishes his own analysis at Growth Memo, he is not employed by any of the companies selling this software, and his reputation rests on whether his methodology survives being picked apart. That is a different kind of source from a marketing page with a multiple on it.
For transparency, the study was run in partnership with AirOps, who do sell AI visibility software, and the webinar where I heard the rest of this was their event. I would rather you hear that from me.
I also think it counts in their favour. Handing your data to an independent researcher, and letting his name and his method carry the findings, is the harder and better way to do this. It is a very different act from publishing your own multiple and asking the reader to take it on faith, because it invites exactly the scrutiny a marketing number is designed to avoid. That is a choice to reward, not to discount.
It holds up on both tests that actually matter.
The methodology is disclosed in detail, down to the embedding model, the similarity thresholds, how many queries were excluded and why, and how many times each query was run. You can argue with it. The conversion multiples give you a number and nothing to check.
And the headline finding is that you have to rank first, which is slow, unglamorous, foundational work. That is the least sellable conclusion available here. A finding that cuts against the easiest pitch has earned more trust than one that flatters it.
So the useful filter is not vendor versus neutral. It is disclosed method versus asserted number, plus whether the conclusion is inconvenient for whoever published it.
"But I don't shop that way"
The most common objection I hear is some version of: I do not use AI to find jewelry, so I doubt my customers do.
Adoption is heavily skewed by income. In a nationally representative survey of 5,031 US adults run by Menlo Ventures with Morning Consult in April 2025, 74% of households earning over $100,000 used AI tools, against 53% of households under $50,000. If you sell engagement rings, heirloom redesigns, or anything with a four-figure price tag, the people most likely to be using these tools are the people most likely to be able to afford you.
The fact that you do not personally search this way is not evidence your buyers do not. They are not you.
You probably cannot see this traffic in your own analytics
Presented at the same session, and worth knowing before you go hunting for proof: roughly 70% of AI-driven traffic is miscategorised in analytics as plain "direct" traffic, and only about 25% of AI search impressions show up in Google Search Console at all.
Direct traffic is the bucket analytics uses when it cannot tell where a visitor came from. So a buyer who spent twenty minutes with ChatGPT, got sent to you by name, and bought a ring can land in your reports looking like somebody who typed your web address from memory.
That has a blunt implication. If you look for this in your dashboard and see nothing, that is weak evidence of absence. The measurement layer has not caught up with the behaviour.
Which leads to advice I would not have given two years ago. For a solo maker, "how did you hear about me" is currently a better data source than your analytics dashboard. Ask it on every enquiry, write the answer down, and you will have a cleaner read on this than any report you can pull. That is not a workaround for a few months. For now, it is the measurement layer.
It is also the strongest argument against precise return figures. Nobody can currently track this cleanly from question to purchase. Anyone handing you an exact multiple is claiming a measurement ability the industry does not yet have.
Both figures come from the talk rather than from a dataset I could check line by line, so treat them as directional. The recording is linked in the sources if you want to see them in context.
What the calculator does, and what it refuses to do
The AI Visibility Calculator is free, needs no email address, and nothing you type leaves your browser.
You set four things: how many extra enquiries a month you want to imagine, what a typical sale from an enquiry is worth to you, the low and high share of enquiries you actually close, and how many months to look at.
It gives you back a range.
It refuses to do several things on purpose, and the refusals are the point:
- It never calls the result expected revenue, or a return, or a projection. The frame is what this would be worth to you if it happened, and the "if" stays yours.
- It will not collapse into a single number. The close-rate control has a low handle and a high handle, and they will not sit on top of each other, because a range with no width is just a point figure wearing a disguise.
- Every number the tool supplies is labelled Assumed and can be changed. Replace it with your own and the label flips to Sourced.
- It never uses a client's result as a default. Nobody else's outcome gets loaded into your model without you choosing it.
Two of those choices make the tool less impressive than the alternative. A single confident figure feels more useful. It is also less true.
What real results look like, with the caveats attached
Bohemi is an alternative engagement ring studio I worked with. Comparing the same months across two years, so seasonal swings do not flatter the story, Nov 2023 to Jun 2024 against Nov 2025 to Jun 2026:
What the work looked like for one client
Bohemi, season-matched: Nov 2023 to Jun 2024 against Nov 2025 to Jun 2026
This was a full rebuild, not an AI visibility retrofit. Product pages, structure, content and measurement all changed together. Attributing these numbers to AI visibility alone would be the same causation error this piece argues against, and they are evidence rather than a setting you should type into a calculator.
Source: Bohemi, first-party, verified against Shopify admin records, 16 June 2026. Bars are scaled for legibility and show relative movement, not a shared axis.
Now the caveats, because they matter more than the numbers.
That was a full rebuild, not an AI visibility retrofit. Product pages, structure, content and measurement all changed together. It would be dishonest of me to hand you those figures and imply AI visibility work alone produced them, and it would be the exact error this whole post is arguing against.
It is also one business, in one category, over one period. Do not put these numbers into the calculator. They are evidence that the work can move a business. They are not a setting.
Frequently asked questions
Do I need to fix my SEO before any of this is worth doing?
In most cases, yes, at least the foundation. If your pages are not indexed, are not readable by machines, or do not clearly describe what you sell, AI visibility work has very little to amplify. The good news is that the foundational work counts twice, because it improves both regular search and AI answers.
Is the calculator going to tell me what I will earn?
No, and it will not pretend to. It shows what a few more enquiries a month would be worth on your own numbers, as a range. Whether you get those enquiries is a judgement you make, not a prediction the tool makes.
How do I tell whether an AI visibility offer is credible?
Ask where the numbers come from. If the return figures were published by the same company selling you the service, treat them as marketing rather than evidence. Ask what they will do to your search foundation, not just to your AI presence.
Sources
- Pew Research Center, "Google users are less likely to click on links when an AI summary appears in the results," published 22 July 2025. Behavioural data from 900 US adults across 68,879 searches in March 2025. Link
- AirOps in partnership with Kevin Indig, "The Fan-Out Effect: What Happens Between a Query and a Citation," published 13 April 2026. 16,851 queries and 353,799 pages across ChatGPT's retrieval pipeline. Position 1 cited 58.4% of the time against 14.2% at position 10; strong heading match 41.0% against 30.2% for weak. Link
- Kevin Indig, "Winners and Losers of AI Search," webinar hosted by AirOps. Source of the traffic-down and conversions-up client case, and of the analytics attribution figures. Presented in a talk rather than in a published dataset, and labelled that way in the body. The client case and the 58% citation figure are stated in the published description of that recording, which is how they were checked for this post. Recording
- AI adoption by household income: Menlo Ventures with Morning Consult, "2025: The State of Consumer AI," April 2025, 5,031 US adults, nationally representative. Link
- Conversion multiples of 4.4x, 5x and 23x: collected 13 August 2026 from marketing material published by companies selling AI visibility services. Deliberately not linked, and cited to show disagreement rather than as evidence.
- Bohemi performance figures: first-party, verified against Shopify admin records, 16 June 2026.
Continue Learning
- Why AI recommends Etsy instead of your jewelry store: the full write-up of the four-engine test behind the verb finding above, including every query I ran.
- How to spot a fake AI visibility expert or tool: the companion to this piece, on judging the person selling you the work rather than the numbers they quote.
- The digital concierge advantage: why interactive tools like this calculator earn citations and rankings at the same time.
- Score your own store on the free AI Visibility Score: the calculator tells you what more enquiries would be worth, and the Score tells you whether you are currently in a position to be named at all.
- Subscribe to the Red Pin Geek Substack (free): weekly inside-view content for jewelry designers building AI visibility on their own. The Premium tier unlocks the working tools, including the Citation Clock Scorecard and the Product Page Audit Scorecard.
- Work with me on your specific store (Readiness Audit $37 / Snapshot $97 / Full Audit $597 / Full Audit + 1:1 call $997): the private read. I run the same foundation check from this piece on your store and tell you what I find.
Want Red Pin Geek in Your AI Answers?
Google now lets you choose preferred sources that get labeled in AI Overviews and AI Mode. If this content helps you navigate AI visibility for your jewelry business, add us and you'll spot our links faster next time Google answers your questions.
Add Red Pin Geek as a Preferred SourceTakes one click. Works in Google Search, AI Overviews, and AI Mode.
