Hi everyone, this is Neo.

There’s been a proper little scrap in the SEO corner of the internet this week, and it started with a blunt LinkedIn post from Ross Hudgens, the CEO of Siege Media: “I recommend everyone remove Perplexity from their LLM trackers today.” His argument: Perplexity’s market share is visibly shrinking, and tracking it with the same weight as ChatGPT and Gemini only distorts your AI visibility data.

The comment section split into two camps. Half applauded and said it was overdue. The other half warned it was “betting today’s leaderboard against tomorrow’s market.”

Honestly, both sides have a point — and both are missing the bigger one. The real question was never “does Perplexity deserve to be tracked?” It’s “what logic should decide who you track, and how much weight each platform gets?” In this post I’ll unpack the genuinely useful stuff hiding behind this argument: why the two big data sources seem to contradict each other, what the market actually looks like after the reshuffle, and the three-tier tracking system I use myself.

1. First, understand the two different “shares” — or the numbers will fight each other

The first reaction most sellers have when they see AI market-share headlines is: why do StatCounter and Similarweb disagree?

Because they measure completely different things:

• StatCounter measures referral share: the share of clicks that millions of websites worldwide actually receive from each AI chatbot. It answers — “who is actually sending traffic to websites?”
• Similarweb measures usage share: the share of traffic to the AI platforms’ own sites (chatgpt.com, gemini.google.com, etc.). It answers — “where do users actually spend their time with AI?”

People constantly blur these two, but they lead to very different conclusions. Here’s the 2026 timeline as it unfolded:

StatCounter’s numbers (share of AI-chatbot referrals to websites): in April 2026 ChatGPT had fallen to 76.85% (from 84.21% a year earlier) but was still miles ahead; Gemini took second at 9.0% for the first time; Perplexity was third at 7.73%; Copilot 3.76%; Claude 2.66%. By June, Gemini (7.94%) and Perplexity (7.91%) were nearly tied. And in the latest August figures cited by Search Engine Journal, Gemini had climbed to 10.9% while Perplexity had dropped to 4.31% — nearly halved in four months.

Similarweb’s numbers (share of worldwide generative-AI web traffic, May 2026): ChatGPT 53.9%, Gemini 27.9%, Claude 9.2%, DeepSeek 4.1%, Grok 2.4%, Perplexity and Copilot 1.3% each. Zoom out over a year and the picture sharpens: ChatGPT’s share fell from roughly 76% to the low 50s, Gemini climbed from under 9% to north of 27%, and Claude went from about 2% to roughly 9% — more than quadrupling in twelve months.

So: Perplexity is declining in both datasets — that part isn’t in dispute. But in Similarweb’s usage data it was only ever at 1.3%, while StatCounter’s referral data still had it at 7.7% in April. If you’re a site owner, referral share is the metric that actually maps to your business.

Lesson one from this whole debate: next time you see a “market share collapse” headline, don’t panic — ask first: which share is it measuring, and does it have anything to do with my business?

2. Was Hudgens right? Yes — but “equal-weight averaging” is the real disease

Hudgens’s concern isn’t baseless. Picture this: an AI visibility tool folds Perplexity into your aggregate score with the same weight as ChatGPT. Perplexity drives a rounding error of the AI traffic in your market — but if you perform unusually well (or badly) there, it can drag the whole score in one direction. One strong or weak result can distort an entire report.

His example makes it concrete: say your brand gets a 40% citation rate on ChatGPT, 35% on Gemini, 30% on Claude, and 90% on Perplexity. A simple average gives you 48.75%. What does that number tell you? Nothing. If Perplexity accounts for 1% of the audience or referrals that matter to your business, why should its 90% outrank the ChatGPT result in your dashboard?

The problem was never “should we track Perplexity” — it’s “why does every platform get the same weight?” Averages have always masked signal. That’s an old disease from the traditional-SEO era of reporting; AI tracking just inherited it.

3. But “delete it” is wrong: the market hasn’t crowned a single winner

Here’s where I think Hudgens’s prescription misses. The 2026 AI search market isn’t “ChatGPT winning while everyone shrinks” — it’s consolidating into a three-way structure where nobody gets optimized away. These three aren’t the same business in three skins; they’re three different distribution machines:

1. ChatGPT: the consumer juggernaut. OpenAI reported 900M+ weekly active users in February and said at the end of July that its models reach more than 1 billion active users across all its products. Scale is its moat — its share is dropping only because the denominator is growing faster; absolute traffic is roughly flat, not shrinking.
2. Gemini: the biggest winner of Google’s distribution. Google said in August that the Gemini app passed 1 billion monthly users. Search, Android, Workspace and Chrome all feed it. Similarweb puts Gemini’s app at roughly 716M MAU in May, up 78% year over year. It’s not winning on product genius — it’s winning on default placement.
3. Claude: the quiet enterprise champion. To consumers it looks small. But Anthropic says 100,000+ customers were running Claude on Amazon Bedrock as of April, told investors in August that annualized revenue has passed $65 billion, and has 1,000+ business customers spending over $1M a year. PwC is rolling out Claude Code and Cowork globally and certifying 30,000 professionals; TCS plans to give 50,000 employees across 56 countries access; Cognizant says 30,000+ associates have completed Claude training.

These are three different distribution advantages, and none of them is going to be “optimized away.” For B2C sellers, ChatGPT and Gemini are the battleground. For B2B sites, SaaS and professional services, Claude matters far more than its consumer numbers suggest.

Which is exactly why a “rank platforms by traffic share and track the top ones” approach is a disaster for B2B: the platform that sends the most referral traffic today is not necessarily the platform that deserves the most of your attention.

4. And the layer most people miss: Google’s “AI search layer” is not just another LLM

There’s a subtler confusion buried in this debate: should Google AI Overviews and AI Mode be treated as “another LLM”?

My answer — same as the original piece — is no. They’re a tier above. Look at the numbers:

• AI Overviews: Google said in June it reaches 2.5 billion users a month;
• AI Mode: over 1 billion monthly users, with query volume more than doubling every quarter since launch;
• Similarweb data reported via TechCrunch: AI Overviews now appear in 43% of Google’s U.S. searches (up from 15% a year earlier); AI Mode visits grew from 126M in June 2025 to 279M in May 2026.

Fold AI Overviews and AI Mode into a weighted average with standalone chatbots and you’re comparing an AI layer living inside the world’s dominant search engine against a few standalone apps. That’s like ranking the search-results page itself against “a browser” — wrong dimension entirely.

They’re not another acquisition channel. They’re Google Search itself going AI-native. For your site, this layer needs to be tracked, optimized and reported on separately.

5. Neo’s practical system: replace “one leaderboard” with three tiers

If a single leaderboard lies to you, what should you do instead? Here’s the system I use — copy it wholesale:

Tier one: platforms with both scale and strategic significance — track separately, never merge

• Track ChatGPT and Gemini as separate rows, don’t fold them into a generic “LLM score” — their distribution logic and audiences are completely different;
• If you do B2B, SaaS, professional services or sell to enterprises, add Claude as its own row. Its consumer traffic looks small; your customers don’t live there — they live in Claude.

Tier two: AI search embedded in existing ecosystems — its own group

• Google AI Overviews and AI Mode get their own group. The goal isn’t measuring an LLM; it’s measuring how AI is changing the search journey itself;
• If your business leans on the Microsoft ecosystem, put Copilot here too — Microsoft says its Copilot family has 150M+ monthly active users and its AI features across all products reach 900M. That overlap with your customer base may be bigger than you think.

Tier three: emerging or niche platforms — watch, but don’t weight

• Perplexity, Grok, DeepSeek live here: don’t give them equal weight, but don’t delete them either. Keep a light eye on the anomalies — citation spikes, referral traffic, growth after new features. If a curve starts bending upward, promote it to tier one later. That’s a promotion path, not an afterthought.

On top of the tiers, connect three datasets instead of staring at one score:

1. Audience exposure: which platforms your actual market spends time on (the “usage share” lens);
2. Brand visibility: how often your brand and links get mentioned and cited per platform, and what prompts trigger them;
3. Business outcomes: whether AI-referred traffic turns into engagement, leads and orders.

And one trap worth naming: a “mention” and a “citation” are not the same thing. Getting named without a link, or being used as a source in an answer that never names you — different events, different business value. Track them separately; don’t let a tool mash them into one number.

Quarterly ritual: run a platform review and answer three questions — ① Which platform’s referral traffic moved more than 20% up or down last quarter, and why? ② Where do your target customers (not everyone) actually spend time? ③ Is a visibility-score change a real signal, or a measurement artifact (model swaps, prompt updates, tool methodology changes all drift scores)? Answer those three and your dashboard organizes itself into tiers.

Neo’s take

Deleting Perplexity is “betting today’s share against tomorrow’s market” — and someone already lost that bet in 2002. The SEJ author told a story I really relate to: back in March 2002, the industry was busy consolidating its reporting around the “five major search engines” — Yahoo, Excite, Lycos, AltaVista, Ask Jeeves. Google wasn’t on the list. Nobody considered it one of the agreed-upon incumbents. And we all know how that ended. The people who missed Google didn’t lose on technique — they lost on assuming the leaderboard was settled.

Perplexity today looks a lot like that “small player”:

• It genuinely is shrinking — StatCounter has it falling from 12.07% in April 2025 to 4.31% in August 2026;
• But it’s still at the table — Similarweb’s August analysis still calls it an active player in AI search, and it even partnered to bring market intelligence into Perplexity Computer. Its ad-free, subscription-plus-enterprise model still resonates with a specific crowd;
• And what if next year some vertical — say, professional Q&A in finance or health — gets eaten up by it? Delete it today and you’ve just blindfolded yourself in that vertical.

My verdict: Hudgens diagnosed the disease correctly and prescribed the wrong medicine. Don’t remove Perplexity from your tracker — de-weight it. If it’s 1% of your AI referrals, giving it 25% of your score is indefensible. If it drives more than 5% of your AI referrals in a specific vertical, it belongs in tier one. The same principle applies to every new AI platform that shows up next.

One last line, which is the biggest lesson years of monitoring have taught me: track the leaders, weight them by reality — but always keep one eye on the outsiders gaining ground. In search, the platform you stop tracking is often the one you later wish you hadn’t.

Markets haven’t settled, so don’t weld your dashboards shut.