Perplexity Now Sends Less US Traffic Than Microsoft Copilot. Your Dealership's AI Visibility Report Is Probably Weighted for Last Year.

Somebody in the SEO business proposed deleting an entire platform this week, and the argument is worth a dealer's attention for reasons that have very little to do with the platform.

Ross Hudgens, CEO of Siege Media, posted that Perplexity's market share has shrunk far enough that including it alongside ChatGPT, Gemini and Claude gives marketers a distorted read on where their visibility actually matters. His words: "Everyone should remove Perplexity from their LLM trackers today." Greg Jarboe took the other side at Search Engine Journal on September 9, writing, "I think Hudgens is asking the right question. I am less convinced that he has the right answer."

Both of them argued it from worldwide numbers. Pull the same dataset for the United States and it gets sharper, and a good deal more useful to anyone selling cars in Ohio.

The US table

Statcounter publishes a monthly AI chatbot share for the United States, and the page will hand you the underlying series as a CSV. Here is what twelve months did to it.

In September 2025, the US ranking was ChatGPT at 78.32%, Microsoft Copilot at 9.54%, Perplexity at 7.72%, Google Gemini at 3.05% and Claude at 1.33%. Perplexity was the third largest standalone AI referrer in the country, and it was sending more than twice what Gemini sent.

By August 2026 the same table reads ChatGPT 72.26%, Google Gemini 11.71%, Microsoft Copilot 7.89%, Claude 4.37% and Perplexity 3.69%.

Perplexity went from third to fifth. Gemini went from fourth to second, nearly quadrupling along the way. Microsoft Copilot slipped one place to third, and it is still sending more than double Perplexity's share.

The Copilot line is the one that should stop a dealer, because Copilot is the platform I almost never hear brought up in a dealer meeting. Perplexity comes up constantly. The two have been moving in opposite directions for a year.

Read the metric before you spend anything against it

Statcounter is measuring referrals. Its FAQ is blunt about what that means: "It's the click to a website that we measure, not the actual search queries that were performed." A referral gets logged when a person clicks through and a page loads with Statcounter's code on it. The company says it calculates on "more than 3 billion page views per month" across "1 million+ member sites," and that "no artificial weightings are used."

So the table answers exactly one question: which assistants actually deliver a human to a website. For a dealer that is the best question on the board, because a click is a shopper standing on a VDP.

It answers a different question than whether an AI mentioned your store. Most AI visibility products sold to dealers are measuring mentions. Referral share is the wrong weight for a mention score unless the vendor tells you that is what they did.

The 2002 problem

Jarboe's rebuttal leans on a story worth repeating in a dealership.

At Search Engine Strategies in Boston in March 2002, somebody proposed simplifying reporting down to the five major engines of the day: Yahoo, Excite, Lycos, AltaVista and Ask Jeeves. Jarboe pointed out that Google was missing from the list.

The engine that was too small to make the cut took the market. That is the case against pruning a platform out of a report because it had two bad months, and this dataset supplies its own argument for the caution.

Claude sat at 7.18% in the US in July and 4.37% in August. ChatGPT ranged from 61.7% in April to 72.26% in August. Perplexity's fall from 5.96% in July to 3.69% in August is a single month of movement inside a series that moves constantly.

Statcounter also reserves the right to change the recent months. Its FAQ says the stats are subject to correction and that "after the expiration of a 45 day period from first publication, no changes will be made to the data." August 2026 is still inside that window.

The artifact that undercuts the headline

Here is the part largely absent from the industry conversation this week, sitting in plain sight in Statcounter's own FAQ.

Asked directly whether it tracks AI browsers, Statcounter answers: "We've researched Comet from Perplexity, Dia from The Browser Company and Chatgpt Atlas from OpenAI. All three browsers are identifying as the Chrome browser. If they identify as their own unique browser in the future, we will track their usage. For now, we can only record their usage under Chrome."

Now read that against the headline number. Perplexity has been pushing its users into Comet. Every referral arriving through Comet lands in the Chrome bucket instead of the Perplexity bucket. The same goes for ChatGPT Atlas. So the measured decline of a platform that ships its own browser is partly a measurement decision, and the direction of the error is knowable: it undercounts.

That does not rescue Perplexity. Third place to fifth is a real move, and the trend across twelve months is consistent. It does mean the confident version of the claim, that Perplexity is finished and can be deleted, rests on an instrument that by its own admission cannot see part of what it is measuring.

What this means for your dealership

Ask how the score is weighted. If a vendor hands your store one AI visibility number, that number is an average across platforms, and the weights decide the answer. Give Perplexity the same vote as ChatGPT in a US market and you have let a platform with 3.69% of referrals move your score as hard as one with 72.26%. A single strong or weak Perplexity result then swings a report you are making budget decisions from. Ask for the platform-by-platform breakdown and the weighting method. If nobody can produce them, the composite number is decoration.

Microsoft is the cheap opportunity in that table. Copilot at 7.89% of US AI referrals is more than double Perplexity's share. Nothing in this data proves what Copilot rewards, and I am not going to invent a mechanism for you. What it does justify is an hour of somebody's time confirming your store name, address, phone, department hours and service listing are correct in Microsoft's own listing product at Bing Places. Boring listing hygiene, aimed at the third largest AI referrer in the country.

ChatGPT is still the whole ballgame among the standalone assistants. At 72.26% of US AI referrals in August it outweighs the other four combined by roughly two and a half to one. We covered the 92% version of this finding back in July, and twelve months of Statcounter data points the same direction.

The biggest AI surface for your store sits outside this table. Google says AI Mode "has surpassed a billion monthly active users globally" and that AI Mode queries "have more than doubled every quarter since launch." Its US breakdown adds that the average AI Mode search runs triple the length of a traditional Search query, and that searches beginning with "where to" and "where should I" are growing. That is the exact shape of a local car shopper's question. The standalone chatbot table is a sideshow beside it, which is why the AI report in your Search Console deserves its own line in your reporting.

Treat volatility as the standing condition. The top of this table reordered twice in twelve months. We have already argued that a single AI visibility reading is mostly noise, and that asking the same question twice changes the answer. This adds a second axis to that: the platform mix underneath the score moves too, and faster than most reporting cycles.

What to do about it

1. Get the weights in writing. Per-platform numbers and the weighting method from whoever produces your AI visibility report. One email.
2. Reweight to your own data. Your GA4 referral traffic for the last twelve months is a better weight for your store than any published national average. Look at what actually arrived.
3. Spend one hour in Bing Places. Name, address, phone, hours by department, service listed and correct.
4. Keep the underlying work platform neutral. Crawlable HTML, current inventory, accurate pricing, real department pages. That same work serves all five assistants and Google's AI surfaces at once.
5. Break Google out from the chatbots. AI Overviews and AI Mode belong on their own line, measured in Search Console rather than folded into an LLM average.
6. Down-weight a fading platform instead of deleting it. Deleting it means you find out late when it comes back, and 2002 is the cautionary tale.

The line to watch

Hudgens is right that equal weighting distorts the picture. Jarboe is right that the fix is a change in weights rather than a deletion. For a dealer, the specific argument matters less than the habit it exposes, which is that most stores are handed AI visibility numbers with no explanation of how they were built, then asked to spend against them.

The platforms will keep trading places. Gemini nearly quadrupled in the US in a year, and it took the number two slot from a company that had twice its share the previous September. Anyone certain about the 2027 ranking is guessing.

If you want a second set of eyes on how your site and your listings look to these systems before Q4 budgets lock, book a walkthrough and we will go through the referral data with you.