Five Insites

The biggest AI categories are the emptiest

By Insites Marketing Team

There's an uncomfortable pattern running through this week's five, which is that most of the deciding now happens somewhere you aren't standing. Half of ChatGPT's categories still have nobody clearly owning them, and the biggest ones are the emptiest of the lot. Only 38% of the pages Google cites in AI Overviews rank in its own top ten. The passages getting quoted over and over turn out to be 117 words long with the answer up front. Pages loading in under 0.4 seconds pull three times the citations, which is a deeply boring reason to win. And 80% of the B2B buying journey now runs with no rep involved at all. Here are your five.

1. Half the categories in ChatGPT still have nobody sitting on them, and the biggest ones are the emptiest

Semrush tracked more than 50,000 brands across 1,094 categories from January to June, five prompts per category, the sort of thing an actual buyer types rather than the sort of thing a keyword tool suggests. Only 15.2% of those categories had a clear owner. Another 31.2% had someone emerging. The remaining 53.7% were a free-for-all with nobody in charge. The part that stings: the biggest categories, the ones carrying 98% of all AI search volume, had a clear owner just 11.3% of the time, against 19% for the sleepy little ones. The more people asking, the less settled the answer.

And the signals you'd instinctively reach for to predict a winner barely function. Branded search volume picked the owner 55.7% of the time. Organic traffic managed 48.4%, which is worse than guessing. Authority Score got 52.5%. You could flip a coin, cancel the subscription and do roughly as well. There is still unclaimed land in exactly the places that matter most, which is either terrifying or the best news you'll read this month, depending on how your week is going. Semrush

2. The paragraph Google quoted 661 times was 117 words long and answered the question in its first sentence

Pillarbase went through 15.7 million AI Mode citations across 148 industries and pulled out 4.6 million distinct passages from 2.7 million pages. Most of them are one-hit wonders: 80.9% were cited exactly once and never again. But around 2,300 passages were reused 61 times or more, and the hardest-working one of the lot got quoted 661 times across 483 different queries. Somebody wrote a single paragraph and it has been on overtime ever since.

The winners share a shape that is almost insultingly learnable. Median length 117 words, so a proper paragraph rather than a clipped snippet. 80% put the answer in the first sentence. 85% made complete sense with nothing around them for context. Passages phrased as a question were recycled twice as often, 48% against 22%. No trick in there anywhere, just the unglamorous discipline of answering the question before you start decorating it. Search Engine Land

3. Only 38% of the pages Google cites in AI Overviews rank in Google's own top ten

Ahrefs went through 863,000 keywords and four million AI Overview URLs and found that just 38% of cited pages appeared in the top ten organic results for the same query. Roughly 31% came from positions 11 to 100. Another 31% came from beyond position 100 entirely, which is to say from pages Google's own ranking system had quietly filed under "no thanks". In their previous run, that top-ten figure was 76%.

Before anyone redraws a strategy deck around a number halving, Ahrefs are refreshingly upfront that they improved their parsing between the two studies, so the datasets aren't directly comparable and some of that drop is measurement catching up rather than the ground moving. The direction still holds: being cited and being ranked are drifting apart, and the query fan-out behind AI Mode pulls sources from places a rank tracker has never once looked. Your rank tracker is now, at best, a partial witness. Search Engine Journal

4. Pages loading in under 0.4 seconds got three times the AI citations, which is a deeply boring reason to win

An analysis of more than eight million sites using Chrome's own user experience data found that pages with a First Contentful Paint under 0.4 seconds averaged 6.7 citations across ChatGPT, Perplexity and Copilot. Slower pages averaged 2.1. Mobile Core Web Vitals pass rates, meanwhile, sit at 48%, up from 32% in 2021, which reads like progress right up until you realise it means more than half of mobile pages still don't pass.

Nobody has confirmed speed is a direct ranking factor in AI search, and the honest reading is that it works as a gate rather than a lever: slow pages get crawled less generously and fall out of the candidate pool before any interesting judgement happens. It is genuinely deflating that after all the frameworks and the positioning workshops, one of the cleanest predictors of getting quoted by a machine is whether your page turns up on time. Analyze

5. 80% of the buying journey now happens without you in the room

Consensus' 2026 buyer report puts the share of the purchase journey involving any direct supplier contact at 17%. Sixty-one percent of B2B buyers say they'd rather buy with no rep involved at all. And 77% described their most recent purchase as highly complex or difficult, which is a remarkable pair of findings to sit side by side: people are finding this genuinely hard, and would still prefer you stayed out of it.

Worth knowing the report comes from a company selling demo automation software, so their product metrics deserve the appropriate squint. The buyer behaviour holds up regardless, and it rhymes with everything above it: whether it's a model summarising you or a buyer reading about you at eleven at night, the decision is being made somewhere you aren't standing. Whatever you wrote down when nobody was watching is doing the selling now. Consensus

Small business shout-out

This week's shout-out goes to Ting Tang, which makes a dry wash shampoo that actually washes. The foam emulsifies the dirt in your hair and you towel it straight back out, so there's no sink, no shower and no water involved at any point. There are separate formulations for adults, teens, tots and pets, and they reckon a 100ml bottle will see you through the best part of a year.

Founder Purdey Collier, a former banker, built it around her daughter and everyone else for whom a normal hair wash is quietly a bit of an ordeal: children with autism or sensory processing differences who can't stand water in their ears or shampoo in their eyes, and people living with chronic illness, limited mobility or fatigue. It has since picked up Theo Paphitis' Small Business Sunday backing, a spot on BBC Radio 4 and a write-up in Stylist. Know a small business we should feature? Reply and let us know.