The July reading, in plain numbers.

Ryan Thompson · Editor · Published July 3, 2026 · Expanded July 6, 2026 · On the July 2026 edition

We build this site to be read by people and by machines, and the two turn out to want the same thing: structure. So this month the note leads with the data itself, laid out as one plain table you can read top to bottom, tab through with a keyboard, or hear read aloud by a screen reader in the order it is written. Every row names its source. Ranges appear where sources genuinely disagree, because the spread is the honest number.

The July reading at a glance

1. The panel partner behind this lineage changed from Datos to Similarweb between editions. Same source family, noted for transparency.
2. This figure comes from a different, wider Semrush study than June's single-search number, so read it as an updated range, not a clean doubling.
3. Reported as Goodie's own monthly trend line, which falls from 72.5% in January to 62.6% in April. First place, narrowing lead.
4. 5W Research independently reports the same collapse, from about 70% down to under 20%. That is new corroboration, not new movement.

How to read it

Two numbers carry most of the month. Zero-click crossed to roughly two in three Google searches, and the monthly reading ticked up to 75, which is our single measure of how hard it has become for a brand to be reached inside an AI answer. Higher is harder. The rest of the table is context around those two.

The engine row is the one people misread most. ChatGPT is still comfortably first in AI referrals, but its share is drifting down as Claude and Gemini pick up ground. That is not a collapse, it is a market widening. Where your customers ask is becoming as real a question as what they ask, and the answer is no longer just one name.

Two in three searches now end on the page

Sixty-eight percent of Google searches ending without a click is the kind of number that reads like an abstraction until you put it against a budget. Most brands still plan as if the search result is a doorway. Increasingly it is the destination. The searcher asks, the engine answers, and the visit that used to follow simply does not happen. Inside Google AI Mode the effect is close to total, at about 93% zero-click.

The instinct is to read that as loss, and in raw traffic terms it is. But the same table carries the counterweight: visitors who do arrive from an AI answer convert at 14.2% against 2.8% for traditional organic. Fewer people click through, and the ones who do have already had the comparison shopping done for them inside the answer. The practical question for a brand is no longer how much traffic search sends. It is whether you are present in the answer that replaced the visit, and whether the smaller stream that still clicks is landing on pages built to receive someone who is nearly decided.

The engine mix is widening, not flipping

ChatGPT still carries most measurable AI referral traffic at 62.6%, with Claude at 18.5%, Gemini at 10.6%, Perplexity at 7.3%, and Copilot at roughly 4%. Two things are true at once in that row. The lead is large, and the lead is shrinking. Goodie's own trend line has ChatGPT falling from 72.5% in January to 62.6% in April, and the share it gives up is being absorbed across the field rather than by one challenger.

For a brand, the operational reading is coverage. Being visible in one engine is no longer the same thing as being visible in AI search, because the engines build answers differently and cite different sources. A page that Claude cites may never surface in Gemini. The panels measuring these shares also disagree with each other more than they disagree month to month, which is why we treat the ordering as more durable than the decimals.

Ranked is not cited, and the gap is now corroborated

The overlap between holding a top-ten Google ranking and being cited in the AI answer for the same query sits at 17 to 38%, down from around 70 to 75% in mid 2025. What changed this month is not the number. It is that 5W Research independently found the same collapse, which moves the finding from one methodology's claim to a corroborated pattern. When two unrelated shops measure the same fall, the fall is the story.

The uncomfortable implication is that a decade of rank-first investment does not automatically transfer. Citation appears to favor pages that answer engines can parse and trust: real text over rendered widgets, named entities, sourced claims, and structure a machine can walk. Ranking is still important. Being mentioned is becoming essential, and the two are now measurably different outcomes.

What we are watching, not predicting

Same discipline as always. We track whether the engine mix keeps widening, whether citation keeps concentrating into fewer sources, and whether the pages that get cited keep skewing toward clean structure: real text, clear entities, sourced claims, semantic markup. Answer engines parse a page much the way assistive technology does, which is why a table like the one above is not a stylistic choice. Pages built to be read by everyone tend to be read by machines. We hold this site to a published accessibility standard for the same reason.

---

The full July reading, with every source and range, is in the current edition. The previous reading is the June 2026 edition. The next edition publishes the first Friday of August, numbers locked the Wednesday prior.