01 The market data
ChatGPT is now the ninth most visited site in America
The place your customers go to ask for a recommendation has moved, and your website was never built to be quoted.
One assistant now draws more American visits than all but eight websites, and it answers the question instead of sending anyone onward.
Greg Jarboe pulled the Semrush Traffic Analytics list for July 2026 and set it against July 2025. ChatGPT.com is up 48.38% year over year and now ranks ninth among all United States websites, with 1.09 billion monthly visits. Bing.com is down 50.43% over the same year, the steepest fall anywhere in the top 20.
Read the second number twice. A search engine shedding half its traffic in twelve months is not really a story about Bing. It is a story about where the asking happens now, and the asking has moved into a chat window that answers rather than a results page that refers you onward.
Semrush models traffic rather than measuring clicks. The direction here matches every other measurement of the past year, so the shape is safe. The precision is not, and anyone quoting 1.09 billion to three decimal places is quoting a model.
What this changes for an ordinary business is smaller and stranger than it sounds. A tax practice that spent ten years earning the top of page one still holds that position. What it has lost is a share of the people who used to travel through it. The page still ranks. Fewer people ever arrive.
What to do about it
Open Google Search Console, the free Google tool that reports how your site does in search, set the range to the last sixteen months, and compare total clicks against total impressions. If impressions held steady while clicks fell, you are being shown and not visited, which is the pattern these numbers describe. That gap is the number worth watching from here, and it takes about five minutes to read.
Source
Search Engine Land, “The big winners and losers in U.S. website traffic over the past year,” by Greg Jarboe, September 4, 2026, using Semrush Traffic Analytics for July 2026 against July 2025 · searchengineland.com
02 The evidence
Other people’s lists produced 86% of one company’s AI mentions
The page you write about yourself counts for far less than the roundups other people publish about your industry.
A practitioner set out to confirm his own advice about publishing your own comparison page, and his second test contradicted it.
Zeeshan Yaseen ran two experiments and logged 775 citation events across both. The second one is the interesting one. It ran from May 30 to June 28, tracked 15 buying-intent search phrases across six AI engines, and followed a software company that started with no AI mentions at all.
Of 437 source mentions in that second experiment, roundup lists published by third parties produced 85.8%. The company’s own comparison page produced 14.0%. Press releases produced 0.2%.
The part worth respecting is what the author did with it. After his first experiment he had recommended that clients publish their own listicles. His second experiment told him an owned listicle is a foundation rather than the engine, and he wrote that down.
An honest reading of your own numbers that contradicts what you sold last quarter is rare, and it is the reason this one is worth more than a louder result. This is also uncomfortable for everyone selling AI visibility, Forever Cited included, because it puts the biggest lever outside the website a client is paying anyone to improve.
Hold it loosely. Two experiments, one practitioner, one client each, no independent replication. A software company with no prior mentions is close to the easiest case there is, and a veterinary hospital or an insurance brokerage may sit in a completely different position. The size of the gap is what makes it worth reading, not the decimal.
What to do about it
Search for the roundups that already exist in your field, the best-of and top-ten pages a stranger would land on, and find out whether you appear on them. Most are maintained by someone reachable. Getting considered for a list you are genuinely eligible for is ordinary business development, and on this evidence it does more than another page about yourself.
Source
Search Engine Land, “Two GEO experiments challenge conventional AI visibility advice,” by Zeeshan Yaseen, September 14, 2026. Experiment two ran May 30 to June 28, 2026, 15 keywords, six platforms, one client · searchengineland.com
03 The practitioner view
Bulk AI content was rated a negative by 131 professionals
If your marketing budget is buying a high volume of AI-written pages, the people who watch rankings closest expect it to cost you.
This is collected opinion rather than measurement, and the opinion is unusually settled on one point.
Zyppy asked 131 search professionals to rate 103 possible ranking factors on a seven-point scale and to name the three they thought mattered most. That produced 13,665 individual ratings.
Relevance came first, named in the top three by 57.1% of respondents. Backlinks came second at 54.8%, which is a quietly remarkable result in a year when most of the conversation has been about everything except links.
AI-generated content published at scale with little added value received negative ratings. Not neutral, and not merely unhelpful. The practitioners rated it as working against the sites that do it.
Now the caveat, and it matters more than the finding. This measures what experienced people believe, not what Google actually does. A survey of opinion cannot separate a real ranking signal from a widely shared assumption, and this field has held confident wrong beliefs for years at a time.
Read it as a warning rather than a finding. An architecture studio publishing forty AI-written city pages is making a bet that the most experienced people in this field think is a losing one. That is worth knowing before the invoice, even though nobody here has proved the mechanism.
What to do about it
If anyone has proposed a content program to you measured in pages per month, ask what happens to the pages that get no visitors after ninety days. A plan that has no answer to that question is a volume plan. Ask for fewer pages, each one answering a question you get asked on the phone.
Source
Search Engine Land, “Google’s top ranking factors, according to 131 SEO professionals,” by Danny Goodwin, September 9, 2026, reporting Zyppy’s 2026 Google Ranking Factors Expert Survey · searchengineland.com
04 The platform test
Google is testing ads that look like the answer
If a paid placement becomes hard to tell from a recommendation, the weight your customers give that answer changes.
This is one practitioner’s screenshots rather than a Google announcement, which is exactly how much weight it deserves.
Sachin Patel spotted text link advertisements inside Google’s AI Mode and posted the screenshots. Barry Schwartz reported them on September 15. The links sit in the body of the response and read like the assistant’s own words, with a Sponsored label placed above the answer.
Google showed plain anchor links inside AI Mode responses back in April. What is new is that these ones are paid, and the paid version is difficult to distinguish from the unpaid version.
Google has said nothing about it, and one person’s screenshots of a test is not a launch. Tests get withdrawn constantly. This one is on the list because of what it would mean if it shipped, not because it is settled.
The disclosure question is the one to sit with. A reader who cannot see where the recommendation stops and the advertisement starts has no way to weigh either. That is a problem for the reader first, and it becomes the advertiser’s problem immediately afterwards.
Your own advertising rules follow you into this channel exactly as they followed you into the last one. A wealth advisor buying placement here is still bound by the Securities and Exchange Commission marketing rule covering testimonials and performance claims, and an insurance broker is still bound by the advertising rules of every state the broker is licensed in, whatever the format looks like on screen. A new surface has never yet come with a new set of rules.
What to do about it
Nothing, this week. Open AI Mode, ask it the question your best customer would ask, and look at what sits above the answer. Knowing what the surface looks like before somebody sells you a placement on it is the whole of the work right now.
Source
Search Engine Roundtable, “Google AI Mode Tests Text Link Ads,” by Barry Schwartz, September 15, 2026, reporting screenshots from Sachin Patel. No statement from Google · seroundtable.com
05 The money
Facebook clicks got cheaper, and real estate gained the most
While everyone watches the AI answer, reaching people on the older channel has quietly become cheaper for most businesses.
The falling price of a click is real, and the report comes from the company that sells the campaign management.
WordStream by LocaliQ looked at nearly 1,800 Facebook advertising campaigns and compared this year against last. For campaigns bought to drive traffic, the average cost per click fell 14.29% to $0.60. The average click-through rate rose 12.87% to 1.93%.
The spread underneath those averages is where the useful part sits. Real estate saw the largest fall, with cost per click down 39.56%. Dental practices were close behind at 41.72% cheaper. Restaurants fell 37.50% and industrial and commercial fell 37.21%.
It did not fall for everyone. Shopping, collectibles and gifts went the other way, up 73.53%. An average across eighteen industries describes none of them, which is the usual trap with a report like this.
The numbers come from a campaign management company reporting on the campaigns it manages. That is not a reason to dismiss them, and it is a reason not to treat them as an industry benchmark. The customers of one platform are not a random sample of anybody.
A cheaper click is not the same as a cheaper customer. If the cost per click in your field dropped by a third this year while the number of enquiries stayed flat, the price went down because the attention did.
What to do about it
Pull your own cost per click for the last twelve months and put it next to your enquiry count for the same period. If the cost fell and the enquiries did not rise, the cheaper clicks are worth less, not more. That single comparison is worth more than any industry average.
Source
Search Engine Land, “Facebook ad costs fall as traffic and lead performance improves: Report,” by Anu Adegbola, September 14, 2026, reporting WordStream by LocaliQ’s analysis of nearly 1,800 campaigns · searchengineland.com
