AI Search Is Taking Your Traffic and Sending Better Customers. Both Are True.

Estimated read time: 11 minutes

The story every small business owner has heard about AI search is that it is eating your traffic. That story is true. The part nobody tells you is that the traffic you are losing and the traffic you are gaining are not the same kind of traffic, and the difference is large enough that some businesses are quietly better off.

We finally have numbers instead of vibes. The agency Brainlabs ran the analysis across 54 of its advertiser clients over a 14 month window starting January 2025. It is not a perfect sample, and we will get to why, but it is real data from real businesses rather than another survey about how people feel about AI.

Here is what it found, what it means for a business your size, and the specific math that determines whether AI search is currently costing you money or making you money.

TL;DR

• Across 54 advertisers over 14 months, organic sessions fell 10.5 percent, from 140.1 million to 125.4 million. 46 of the 54 lost traffic.
• Referrals from AI platforms rose 163 percent in the same period, and conversions from that traffic rose 335 percent.
• Visitors arriving from ChatGPT, Copilot, Gemini and Perplexity completed key actions at 1.5 times the rate of visitors from organic search.
• AI referral volume is still small relative to organic, so most businesses are net down on sessions and net closer to flat on outcomes.
• The businesses hurt worst were the ones whose traffic came from questions an AI answer can fully resolve. The businesses helped most sell something that requires a decision.
• The correct response is not to publish more. It is to publish things an answer cannot replace, and to start measuring conversions per visitor instead of visitors.

What the data actually says

Brainlabs compared 54 advertiser clients across a 14 month window beginning in January 2025. The headline findings:

• Organic sessions fell 10.5 percent, from 140.1 million to 125.4 million across the sample.
• 46 of 54 clients lost organic traffic. That is 85 percent of the sample. This is not a few outliers dragging an average.
• AI platform referrals rose 163 percent.
• AI driven key events rose 335 percent, meaning conversions from that traffic grew roughly twice as fast as the traffic itself.
• ChatGPT referrals alone went from about 645,000 monthly to 2.6 million between June 2025 and June 2026, growth of 303 percent.
• AI referred visitors triggered key events at 1.5 times the rate of organic search visitors.

The steepest organic declines showed up in fitness, financial technology, insurance, and consumer packaged goods. Hold that list, because the pattern in it explains most of what follows.

Two honest caveats. First, these are agency clients, which means they are businesses already spending money on marketing and are probably larger and better optimized than a typical small business. Second, 14 months is a short window during which Google was aggressively rolling out AI Overviews and AI Mode, so some of the decline is a product change rather than a permanent shift in how people search. Neither caveat changes the direction. Both should make you cautious about the exact percentages.

Why 10.5 percent is smaller and worse than it sounds

A 10.5 percent decline over 14 months is milder than the apocalypse a lot of the SEO industry has been selling. If you had been told your organic traffic was going to halve, and it fell by a tenth, that is survivable.

It is also worse than it sounds, for a reason that gets lost in the aggregate: the decline is not evenly distributed across your pages. It is concentrated almost entirely in content that answers a question completely.

Think about what an AI answer can fully resolve. What is a good resting heart rate. How does a Roth conversion work. What does comprehensive coverage mean. How many ounces in a cup. These are questions with a correct answer that fits in a paragraph, and a generated answer serves them perfectly well. Nobody needs to click through to a 1,800 word article that buries the answer under an anecdote about the author’s grandmother.

Now look at that industry list again. Fitness, fintech, insurance, and packaged goods are all sectors whose content marketing consists overwhelmingly of explainer articles answering exactly those kinds of questions. They were not unlucky. They were built on the one traffic source that AI search directly replaces.

So the real number is not 10.5 percent across the board. For most sites it is closer to a much larger decline on informational pages and roughly flat performance on pages where someone is comparing, deciding, or buying. If your blog is 80 percent explainers, your decline will be steeper than the average. If your traffic already came from comparison and purchase intent pages, you may barely have noticed.

We covered the mechanics of this shift when Google restructured the search results page itself. This data is what that restructuring looks like fourteen months into people’s analytics.

The 1.5x, and why AI visitors convert better

Visitors arriving from an AI assistant completed key actions at 1.5 times the rate of organic search visitors. That is a large gap, and the explanation is not that AI users are a better class of person. It is selection.

When someone searches Google, they get ten links and click a few to figure out what they are dealing with. Most of those clicks are orientation, not intent. When someone asks an AI assistant a question, the assistant does the orientation for them. By the time the assistant hands them a link to your site, the browsing phase is already over. They have been told what you are, why you are relevant, and what they should look at. The click that reaches you is the click after the decision, not the click that starts it.

This has a second implication people miss. You are no longer competing for attention on that visit. You are competing for it upstream, inside a conversation you cannot see, where the assistant decided whether to mention you at all. The visit is the result of a competition that already concluded.

It also means the 1.5x is not a permanent property of AI traffic. It is a property of AI traffic right now, while volume is low and the assistants are conservative about which sites they cite. As AI referrals scale, expect the conversion premium to compress toward normal, the same way every early channel premium eventually has.

The math that decides if this is good or bad for you

Forget the industry averages. Here is the calculation for your business, and it takes about ten minutes.

You need four numbers from the last twelve months: organic sessions, organic conversions, AI referral sessions, and AI referral conversions. A conversion here means whatever actually matters to you, which for most small businesses is a form fill, a booking, a call, or a sale. Not a newsletter signup you never follow up on.

Then run this comparison:

1. Conversions lost to organic decline. Multiply the sessions you lost by your organic conversion rate.
2. Conversions gained from AI referrals. Multiply your AI referral sessions by your AI referral conversion rate.
3. Subtract.

A worked example. Say you dropped from 10,000 to 9,000 monthly organic sessions, and your organic conversion rate is 2 percent. You lost 20 conversions a month. Meanwhile AI referrals went from 50 to 400 sessions at a 3 percent rate, so you gained about 10.5 conversions. Net: you are down roughly 9.5 conversions a month and AI referrals covered about half the loss.

That is the typical shape right now for a business with a healthy content library. The volume gap is real. The gap is closing faster than the loss is growing, because AI referral conversions in the Brainlabs sample grew 335 percent while organic sessions fell 10.5 percent.

Run the number before you panic and before you celebrate. Businesses selling considered purchases, professional services, and anything where people compare options tend to come out better than the average. Businesses whose model was built on ad impressions against high volume informational traffic tend to come out much worse, because they lose the impressions and the smaller, better traffic does not replace ad revenue the way it replaces leads.

What this changes about what you publish

The reflex advice is to publish more. That is exactly backwards. If informational content is the category being absorbed by AI answers, publishing more of it is spending money to produce the thing that no longer returns traffic.

What holds up is content an answer cannot substitute for. In practice that is four things.

Things only you know

Your own pricing, your own results, your own process, your own data. An assistant can summarize what the internet already says about invoice factoring. It cannot tell someone what you charge, what your turnaround actually is, or what happened with the last eleven clients in their situation. Proprietary information is the only content with no substitute.

Judgment on close calls

Generated answers are good at consensus and bad at taking a position. Content that says “most people recommend X, and here is the specific situation where that is wrong” performs well, both with readers and with the assistants themselves, which tend to cite sources that add something rather than restate the average.

Comparison and decision pages

These held up better than explainers because the person reading them is deciding, and deciding is the one thing people still want to do themselves. They also happen to be the pages that convert, which is why this is the least painful adjustment on the list.

Structure an assistant can read

If you want to be the source an assistant cites, make the facts legible: clear headings that match real questions, specific numbers rather than ranges, dates on your claims, and schema markup so a machine can parse your prices, hours, and offerings without guessing. None of this is new advice. It is just that the payoff moved from a rich snippet to a citation.

One thing that does not work: writing for the assistants instead of for people. The output reads like a wiki entry, converts badly, and the assistants are not fooled by it either. Write the thing a competent human would find useful and make it structurally easy to parse.

How to measure it without buying anything

You do not need a new platform for this. You need to look at data you already have, segmented in a way you probably have not tried.

Isolate AI referral traffic

In your analytics, build a segment for referral sources containing chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com, and claude.ai. That segment is your AI channel. Most analytics tools bury these in generic referral traffic, which is why so many businesses believe they get zero AI traffic. They usually get some, and it usually grew.

Split your pages into informational and decision

Tag each significant page as one or the other, then compare traffic trends for the two groups over the last year. This single view tells you more about your exposure than any industry benchmark will. If your informational pages are down 30 percent and your comparison pages are flat, you know exactly where to stop investing.

Switch your reporting metric

If you report on sessions, you are going to spend the next two years watching a number go down while your business is fine. Report conversions, and conversions per thousand sessions. The first tells you whether you are making money. The second tells you whether the traffic you get is getting better or worse, which is the actual question this era poses.

If your current analytics setup makes any of this painful, that is worth fixing before you spend on anything else. We compared the practical options in website analytics tools for small business.

What not to change

Do not delete your informational content. It is still what gets you cited. An assistant that has never seen your site is not going to recommend you, and the explainer that no longer earns clicks may well be the reason your name appears in an answer. Its job changed from generating traffic to generating credibility. Stop measuring it by clicks, but leave it up and keep it accurate.

Do not abandon traditional SEO. Organic search still delivered 125.4 million sessions in a sample where everyone is convinced it is dying. It shrank by a tenth. It did not disappear, and the fundamentals that earn rankings are the same fundamentals that earn citations. The tooling question has not really changed either, which is why the standard SEO tool stack still does most of the work you need.

Do not buy an AI visibility product on the strength of a scary chart. Some of these tools are useful and some are repackaged rank tracking with a new label. We went through which is which in our look at AI SEO tools and the snake oil among them. Do the free measurement first so you know what problem you are actually buying a solution to.

Do not chase every platform. ChatGPT referrals grew from 645,000 to 2.6 million monthly in this sample while other assistants stayed comparatively small. Your attention has a cost. Concentrate it where the volume is until that changes.

Frequently asked questions

Is a 10.5 percent organic decline what I should expect?

Only if your content mix resembles the sample’s. The number is an average across 54 businesses, 46 of which declined. If most of your traffic comes from questions an AI answer resolves completely, expect worse. If most of it comes from comparison, local, or branded searches, expect better. Segment your own pages before you adopt anyone else’s number.

Why do AI referrals convert 1.5 times better?

Because the assistant already did the comparison shopping. The visitor arrives having been told what you are and why you are relevant, which means they skipped the evaluation stage that produces most of the bounces in organic search. The trade is fewer visitors who are further along.

I get almost no AI referral traffic. Am I doing something wrong?

Check that you are actually measuring it before assuming you have none, since most analytics setups file these sources under generic referrals. If the number really is near zero, the usual causes are that your site has thin content on the topics you want to be cited for, or that your key facts are not structured in a way a machine can extract confidently.

Should I block AI crawlers to protect my content?

That is a real decision with a real trade, and it depends on your model. If your business is selling access to the content itself, blocking has a logic to it. If your content exists to generate leads or sales for something else, blocking removes you from the answers where your competitors will appear. You cannot be cited by a system that cannot read you.

Will the 1.5x conversion advantage last?

Probably not at that level. Early channels almost always show inflated quality because the users are self selected and the volume is small. Plan on the advantage compressing as AI referrals scale, and treat today’s premium as a window rather than a permanent structural feature.

What is the single highest value change to make this quarter?

Change your reporting metric from sessions to conversions per thousand sessions, and segment AI referrals as their own channel. Almost every bad decision people are making right now comes from watching the wrong number go down.

Related Coverage

• Google Just Killed the Search Box You’ve Known for 25 Years. The product change that produced the traffic pattern in this data.
• Best AI SEO Tools for Small Business (And Which Ones Are Snake Oil). What to buy, and what to ignore, once you know your actual exposure.
• Best Website Analytics Tools for Small Business. The setup you need to segment AI referrals and report conversions instead of sessions.

Faceted Media Magazine covers business, AI, and entrepreneurship for the people building what’s next.