You Still Rank #1. Your Traffic Is Down. Both Are True.
It is the most confusing conversation in small-business marketing right now. Your rankings held. Your traffic fell. Somebody is supposed to explain that, and everybody's explanation sounds self-serving.
Two pieces of news from September 2026 make it easier to talk about honestly. Researchers put a careful number on what AI answers cost one very large website, and Google acknowledged that its own reporting cannot give you position inside an AI answer — with documentation showing the impression counting is loose too.
Neither gives you a number for your business. What they give you is a way to tell a real problem from a structural change — and a short list of things worth measuring once you accept that sessions are no longer the scoreboard.
Would you rather have someone look at your site than read about somebody else's? Book a free strategy call. Otherwise, keep going.
What the Wikipedia Research Found
Two researchers at the University of Washington, Mehrzad Khosravi and Hema Yoganarasimhan, built the cleanest natural experiment anyone has managed on this question.
The setup is clever. AI Overviews launched by default in the United States in May 2024, on English results. German and French Wikipedia did not get the same treatment in the same window. So the researchers matched articles across languages — 499,927 English-German pairs and 530,873 English-French pairs — and compared what happened to search referrals on each side after May 2024, using monthly clickstream data from December 2023 through December 2024.
Their estimate: AI Overviews reduced monthly external search referrals to English Wikipedia by 5.45% relative to German and 4.82% relative to French. In absolute terms, roughly 100 million fewer referrals a month, about 1.2 billion a year.
Separately, the Wikimedia Foundation has reported that human pageviews across all Wikipedia languages fell about 8% against the same period in 2024. Different measure, same direction. Search Engine Journal has the full write-up.
So: the effect is real, it is measurable, and it is smaller than the panic. A 5% referral decline is a serious number at Wikipedia's scale and it is not the extinction event the louder posts have been describing.
What It Did Not Establish
This part matters more than the headline, and I want to be careful here because several write-ups have called this study peer-reviewed. It is not.
It is a working paper. That means the authors have published their method and results for scrutiny, not that other researchers have reviewed and signed off on them. Working papers are how good research gets tested, and plenty of them hold up. But "working paper" and "peer-reviewed" are not the same claim, and anybody telling you otherwise is either careless or leaning on the word to make a point.
Three more limits worth knowing:
- Google disputes the method. Its objection is specific and reasonable: Wikipedia's data lumps all external search engines together, so the analysis cannot isolate Google from Bing, DuckDuckGo, or anything else. That does not make the finding wrong, but it does mean the effect is attributed to AI Overviews rather than proven to come from them.
- Other 2024 events cannot be ruled out. The authors say so themselves. Anything that affected English-language search but not German or French during that window would show up in these results looking exactly like AI Overviews.
- Wikipedia is a strange test case. It is the single most-cited source in AI answers, it sells nothing, and nobody searches for "Wikipedia" when they want to hire a contractor. The kind of query that sends people to Wikipedia is precisely the kind an AI answer replaces best. Your service pages are not in that category.
Which is the honest version of the takeaway: this is good evidence that AI answers take clicks from informational pages. It is not a percentage you can apply to your business, and anyone quoting "5%" as what AI Overviews will cost you is doing arithmetic on a number that was never about you.
Google Says It Cannot Report This Precisely — and It Is Right
Here is the second piece, and it is the one that should change how you read your own reports.
Google's John Mueller, responding on Reddit to a complaint about Search Console's AI metrics, acknowledged the limit directly. "Position for these is hard to do in a way that makes it useful, so we're currently tracking it like we do for many search features (as a block)," he wrote, adding that "the old 'position 1 – 10' is hard to map, or to make useful for site owners." Search Engine Journal covered the exchange.
Translated: when your link appears inside an AI Overview, Search Console gives it the position of the whole AI block, not its place within it. Being the first source cited and the seventh look identical in your report.
The impression counting is looser still. Per Google's own documentation for the Generative AI performance report, which rolled out worldwide on August 31, 2026:
- An impression counts when your link is shown in the AI feature, whether or not the person ever looked at it.
- Links tucked behind a "Show More" expansion are not counted until someone expands it.
- AI Overviews and AI Mode are combined into one number you cannot split apart.
- The data is drawn from the Web search type, so it is not cleanly separable from ordinary search either.
None of that is a scandal. It is Google being straight about a measurement problem it has not solved. But it has a practical consequence: if your AI visibility reporting looks precise, the precision is decoration. Directionally useful is the most any of it can honestly claim right now, and a report that promises you more than that is selling confidence it does not have.
Is Yours a Real Problem or a Structural One?
Rankings holding while traffic falls has two very different causes, and they call for opposite responses. Three checks separate them.
Did Leads Move, or Only Sessions?
This is the first question and most of the time it is the only one that matters. If sessions fell noticeably and your leads, calls, and bookings are flat, you lost the visitors who were never going to contact you — the ones who wanted a fact and now get it in the answer box. That is a structural change, and it is not a problem.
If leads fell with sessions, stop reading about AI and go find out what broke. That is a real problem and it usually has an ordinary cause.
Which Pages Lost the Traffic?
Sort your losses by page type. Blog posts and informational pages are the ones AI answers replace — "how long does a roof last," "what is the difference between X and Y." Service pages, pricing pages, and location pages are not, because somebody researching a purchase still wants to see who they would be buying from.
Losses concentrated in informational pages read as the structural story. Losses on your money pages do not, and they deserve a real investigation.
Did Branded Search Hold?
If people are still searching for your business by name at the same rate, your visibility in the market is intact and the change is in how people get their answers. If branded search fell too, something bigger is going on and AI Overviews are not the headline.
Not sure which one you are looking at?
Send two things to saguarosocialmediaaz@gmail.com: the traffic chart that worried you, and whether your leads or calls moved in the same period. I will tell you which of the two stories you are in, and what I would check first. One paragraph back, no pitch, and no follow-up unless you ask.
What happens to it: I read it, I reply, I delete it. It does not go into a CRM, a mailing list, or a case study.
What a chart cannot tell me is which page or which change caused it. That takes an account, and that is what a call is for.
What to Measure Instead of Sessions
Sessions were always a proxy. They were a decent one when a search result was ten blue links and a click was the only way through. They are a worse one now, and building a reporting habit on a broken proxy is how people end up panicking about a business that is fine.
Four numbers that still mean something:
- Leads, calls, and bookings. The only numbers that pay you. If these hold while sessions fall, you are fine, and any report that leads with sessions is burying that.
- Cost per lead across every channel. Organic search is one source among several. If organic gets more expensive in effort while paid stays steady, that is a budget conversation, not a crisis.
- Branded search volume. The closest thing to a market-awareness gauge you can get for free. Rising branded search on falling sessions is a good sign, not a bad one.
- Conversion rate on the traffic you still get. Fewer, more serious visitors should convert at a higher rate. If they do, the traffic you lost was not worth much. If your rate fell too, the problem is on the page.
This is the same discipline as telling a reporting bug from a real traffic drop: decide what number actually represents money before you decide whether to be worried.
What Actually Helps
Less than the AI-visibility industry would like you to believe, and more than nothing.
Be the Source Worth Citing
AI answers are assembled from pages that say something specific and verifiable. Original numbers from your own business, clear direct answers to real questions, and pages that state plainly what you do and where you do it get cited. Pages assembled from the same seven facts every competitor published do not. That has been the argument for good content for years; AI answers just raised the penalty for ignoring it.
Fix the Inputs You Own
Your Google Business Profile, your service pages, your reviews, and your listings are what AI tools read when somebody asks about a business like yours. They are also the cheapest thing on this list to fix and the only part of it you fully control. Visibility you borrow can vanish without explanation — a lesson Reddit learned expensively this summer.
Do Not Buy an AI Visibility Score
A growing number of tools will sell you a number for how visible you are in AI search. Ask how it is calculated. Google itself cannot break AI Overviews out from AI Mode, cannot tell you where in a block you appeared, and counts impressions for links nobody expanded. Any vendor claiming a precise figure is estimating on top of data Google has publicly called inadequate, and charging you for the confidence.
Check what AI tools say about you — that is genuinely worth doing, and it takes ten minutes of asking a few of them about your business. Just do not pay for a dashboard that turns guesswork into a decimal point.
For the record, I do this work. I check what the AI tools actually say about a client and fix the listings, pages, and reviews they are reading. What I will not hand you is a score. The fix list is real; the number on top of it would be invented.
Keep Paid Search Honest
If organic clicks are getting scarcer for informational queries, the value of showing up on the queries with buying intent goes up, not down. That is an argument for spending deliberately on a small number of high-intent terms, not for spending more. Pick one platform and do it properly before you widen anything.
Common Questions About AI Overviews and Traffic
Five worth answering directly.
Is My SEO Agency to Blame?
Not automatically. A page can hold position one and lose clicks because an AI answer now sits above it. The fair question is not why traffic fell but what they changed in response, and whether leads moved or only sessions did.
How Much Traffic Do AI Overviews Actually Take?
Nobody has a number that applies to your site. The most careful estimate so far — a University of Washington working paper that has not been peer reviewed — puts it near 5% of search referrals to English Wikipedia. Wikipedia is an unusual case and Google disputes the method, so treat it as one data point, not a rate.
Can I See AI Overviews Performance in Search Console?
Partly. The Generative AI performance report rolled out worldwide on August 31, 2026. It combines AI Overviews and AI Mode into one figure you cannot separate, counts impressions for links whether or not anyone looked, and skips links behind "Show More" until they are expanded.
Why Does Search Console Show a Position for an AI Overview?
Because Google tracks the whole AI block as one unit. Mueller has said position for these is hard to make useful, so it is tracked as a block like other search features. Your link gets the block's position, not its own.
What Should I Measure Instead of Sessions?
Leads, calls, form fills, bookings, and branded search. Steady leads on falling sessions means the visitors you lost were not buyers. Leads falling with sessions is a real problem worth chasing.