Should You Still Optimize Only for ChatGPT?

No. Optimizing only for ChatGPT used to be a defensible shortcut. It no longer is. ChatGPT's share of measurable AI referral traffic dropped from 89.1% to 62.6% in under a year, while Claude climbed from 1.4% to 18.5%. The traffic is spreading across four engines, and a single-engine plan now misses about a third of it.

That number comes from the Goodie 2026 AI Search Traffic Report, which measured 2,802,519 AI referral sessions across 41 brand sites in its first wave and re-measured the same panel a year later. The shift is not subtle. It is the single cleanest data point we have for why a one-engine strategy is now a liability.

This post does two things. First it shows what the referral data actually says. Then it gives you the per-engine plan that replaces "we'll just win ChatGPT."

What the AI referral data actually shows

Goodie ran a GA4 brand panel, then triangulated it with Similarweb (25.77 billion platform visits, January to April 2026) and SensorTower iOS rankings. Each brand's share of AI referrals was averaged so a single large site could not dominate the result. The headline: the source mix for AI-driven visitors is de-concentrating fast.

ChatGPT fell from 89% to 63% of measurable AI referrals

A year ago, nine out of every ten AI-referred visitors arrived from ChatGPT. Today it is closer to six. ChatGPT is still the largest single source, so it stays in the plan. But "largest" and "only" are different words, and most B2B GEO budgets were built on the second one.

The drop does not mean ChatGPT is shrinking. It means everyone else grew faster.

Claude surged from 1.4% to 18.5% and is now the #2 source

This is the line that should reset your roadmap. Claude went from a rounding error to the second-largest AI referral source in twelve months, a 17.2-point gain. Anthropic's distribution inside SAP, Microsoft 365, and AWS is part of the story, and Claude's referral traffic now outweighs Gemini, Perplexity, and Copilot individually.

If your AI search plan does not name Claude, it is missing your fastest-growing referral source.

We made the case for treating Claude as a first-class target back when it was still a contrarian call, in which LLM you should optimize for. The Goodie data turns that argument from a hunch into a number.

The Big Four now control about 99% of measurable AI referrals

ChatGPT, Claude, Gemini, and Perplexity together account for roughly 99% of measurable AI referral sessions. That is the good news hiding in the de-concentration story: the target list is short. You are not chasing fifty engines. You are covering four, in a known order of size.

The bad news is that those four barely agree on anything underneath the surface.

Why ChatGPT-only optimization quietly fails

A single-engine plan does not fail loudly. Your ChatGPT citations hold, your dashboard looks fine, and your pipeline from AI search just stops growing. Here are the five reasons the gap forms.

Reason 1: The four engines cite almost-disjoint source sets

A cross-platform analysis of 680 million citations, summarized in a 23-study meta-analysis on AI citations, found that only 11% of domains are cited by both ChatGPT and Perplexity. The engines trust nearly separate slices of the web. Winning the sources ChatGPT reads tells you almost nothing about the sources Claude or Perplexity read.

Reason 2: ChatGPT barely cites brands in the first place

The same meta-analysis reports a 46x brand-citation gap between platforms: in a 34,234-response study, ChatGPT named a brand 0.59% of the time versus Perplexity's 13.05%. If you tune everything to ChatGPT, you are optimizing for the engine that is least likely to say your name out loud.

Reason 3: Per-engine source mixes move month to month

MediaPost reported in May 2026 that Perplexity's Reddit citations collapsed from 25% in February to 7% in April, while YouTube in Google AI Mode rose more than fourfold since January. A plan tuned to last quarter's mix is already stale this quarter.

Reason 4: Your fastest-growing referral source is the one you are ignoring

Claude added 17 points of referral share in a year. A ChatGPT-only plan treats that growth as someone else's problem. It is not. It is the part of your AI funnel compounding fastest, and it is sitting outside your measurement.

Reason 5: Referral share is de-concentrating, not consolidating

The comforting story is that one engine will win and you can plan around it. The data says the opposite. Share is spreading out, not collapsing into a winner. A bet on consolidation is a bet against the trend line.

The engines barely overlap, the mix moves monthly, and the leader is shrinking. A single-engine plan is a bet against every one of those facts.

Here is the same split in plain terms.

A ChatGPT-only plan asks:

• •Are we cited in ChatGPT for our core prompts?
• •Did our ChatGPT citation rate go up this month?
• •Which pages does ChatGPT pull from?

A multi-engine plan asks:

• •What is our citation share in each of the Big Four, separately?
• •Which engine are we weakest in relative to competitors?
• •Which sources does each engine read for our category this quarter?

Each question in the second column is a self-contained unit of work. That is the plan.

How to optimize across all four engines

The fix is not more effort on ChatGPT. It is a four-engine plan sized to where the traffic actually is. Work it in this order.

Audit your citation share in each engine separately

Run the same 40 to 50 buyer prompts through ChatGPT, Claude, Gemini, and Perplexity. Record whether your brand appears, and which sources each engine cited, one engine at a time. Roll nothing up into a single average. The average hides the exact gap you are trying to find. We walk through the mechanic in how to run an AI visibility audit.

Fix brand presence before you chase sources

If your brand is missing from two of the four engines, you have a brand-recognition problem, and no amount of source placement fixes it. The repair is named-vendor positioning in analyst content, customer reviews, and industry press, the brand-evidence layer every engine consumes. The brand-versus-source split is the upstream budget decision here.

Build a separate source list per engine

Once you know which engine you lose, map the 25 to 30 domains that engine actually cites for your category. That is your target list for that engine, and it will barely overlap with the others. Place evidence at those sources rather than only on your own site. ChatGPT and Claude lean encyclopedic and authority-weighted; Perplexity leans community; Google AI Overviews leans video and multimodal.

Weight your effort by referral share, not by habit

Size the work to the Goodie order: ChatGPT first because it is still largest, Claude second because it is now genuinely big and growing fastest, then Gemini and Perplexity. Do not spend 90% of the budget on the one engine you already know just because it is familiar.

Re-measure quarterly, because the mix moves

The MediaPost data shows source mixes swinging within a single quarter. Re-run the audit every 90 days and adjust the per-engine source lists. A GEO plan is a subscription, not a one-time install. We cover the cadence in how to measure GEO and AI visibility.

Spend on brand presence first, per-engine sources second, and re-check every quarter. That order survives the next data shift.

What this looks like as a budget split

The de-concentration data does not mean four times the work. It means the same work, pointed in the right proportions.

The point of the table is the proportion. A plan that puts everything into the 62.6% column and nothing into the 18.5% column is leaving its fastest-growing channel unmanaged. We documented why the engines diverge this much in why ChatGPT cites 5 sources but Claude cites 13.

FAQ

Is ChatGPT still the most important AI engine to optimize for?

Yes, it is still the largest single source of measurable AI referrals at 62.6% per the Goodie 2026 report. But "most important" is not "only." Its share fell 26 points in a year while Claude rose 17. Treat ChatGPT as the biggest line in a four-engine plan, not the whole plan.

Why is Claude growing so fast as a referral source?

Claude's referral share rose from 1.4% to 18.5% in a year, making it the #2 source. Anthropic's distribution inside enterprise software like SAP, Microsoft 365, and AWS puts Claude in front of B2B buyers during work, and its citation behavior favors established, authoritative sources, which suits B2B content.

Does optimizing for ChatGPT also cover the other engines?

Mostly no. A cross-platform analysis of 680 million citations found only 11% of domains are cited by both ChatGPT and Perplexity. The engines read nearly separate source pools, so source-level work rarely transfers. Brand-recognition work transfers better than source-placement work.

How often should I re-check my AI search visibility?

Every 90 days. MediaPost found per-engine source mixes shifting within a single quarter, such as Perplexity's Reddit citations falling from 25% to 7% in two months. A plan tuned to last quarter's mix is already out of date, so treat the audit as a recurring task.

What is the fastest way to start a multi-engine plan?

Run your top 40 to 50 buyer prompts through ChatGPT, Claude, Gemini, and Perplexity separately and record where your brand appears. The engine where you are weakest relative to competitors is where the next pipeline is hiding. That single audit tells you where to spend before you change anything.

What to do this quarter

Three moves, in order.

Run the four-engine audit this week. Same prompts, four engines, no rolled-up average. The number you want is your citation share in each engine, separately.

Re-weight the budget by next month. Keep ChatGPT as the largest line, add a real Claude line, and stop pretending Gemini and Perplexity are rounding errors.

Put the audit on a 90-day repeat. The Goodie data shows where the traffic went this year. The MediaPost data shows it will move again. A plan that does not re-measure is planning for a market that no longer exists.

ChatGPT did not get worse. The market around it got wider. The fix is not more ChatGPT. It is a plan as wide as the traffic.