In the first half of 2026, search behavior shifted under our feet, software valuations cratered on sentiment rather than proof, and companies blamed layoffs on AI long before anyone could show the receipts.
Kevin Indig posted a link on LinkedIn last week to his AI Halftime Report, H1 2026, and it is worth noting that his H1 2025 report predicted both Google’s continued AI Mode rollout and the idea that AI layoffs were mostly a PR narrative rather than an operational reality. Both predictions held up, and the report opens with a line that doubles as the thesis for the entire first half of the year: AI’s impact kept growing faster than anyone’s ability to measure it. That gap between impact and measurement is the real story of H1 2026, more than any single product launch or earnings call.
Trust Became A Ranking Factor Readers Feel In Their Own Results
Indig’s research found that roughly three out of four consumers pick the top result in an AI shortlist, unless a brand they already trust appears anywhere else on that list, in which case they pick the trusted name instead. Software stocks fell by close to 30% over the period, and the decline tracked almost entirely with how the market perceived a company’s exposure to AI disruption rather than with how that company actually performed. The bottom quartile of software stocks dragged the whole sector down while the median and top quartile outperformed the broader ETF, which tells you the selloff was a story about narrative, not fundamentals.
Meta engineers reportedly burned through 73.7 trillion tokens in a single month chasing an internal leaderboard that ranked more than 85,000 employees by token consumption. Nobody could point to the return on that spending, and Meta shut the leaderboard down in April once CFOs realized annual token budgets had been blown through in four months.
AI Took The Blame For Cuts It Did Not Make
Layoffs told a similar story of narrative racing ahead of proof. Challenger, Gray & Christmas found AI cited as the reason behind more than 87,000 job cuts through May, roughly a fifth of all 2026 layoffs to that point. Indig’s own reporting, going back to his H1 2025 predictions, argues AI is mostly a convenient explanation for cuts that were really about pandemic over-hiring and capital expenditure discipline. Some of the same companies that blamed AI for layoffs turned around and quietly rehired.
My take (and I have been saying versions of this since spring) is that Indig has named something structural rather than seasonal. Every one of his H1 storylines, the SaaS selloff, the AI layoff narrative, the traffic collapse publishers are now fighting in court, comes back to the same root cause. We built our measurement scoreboards for classic search years ago, and none of them were designed to display something as nonlinear and probabilistic as AI search has turned out to be.
That nonlinear narrative shows up starkly in Indig’s citation data. He found that 91% of citations appear in only one of ChatGPT, Perplexity, or AI Overviews, never in more than one. Prompt tracking, he argues, needs to behave more like polling and focus-group research than like the rank tracking SEOs have leaned on for two decades. Brand mentions, not citations, correlate more closely with real business outcomes, since most buyers care whether a brand shows up favorably across a panel of prompts rather than whether one specific citation landed in one specific answer. Google’s own Search Console data is reportedly 75% incomplete for this new landscape, so even practitioners who think they are measuring carefully are working from a partial picture.
There was an upside to all that token burning. Indig points out that the wave of usage around Claude’s Opus releases pushed far more people into daily AI use than any marketing campaign could have, and that surge fed growth at infrastructure companies further down the stack, the kind of second-order effect that rarely shows up in a quarterly earnings call but shapes an entire ecosystem anyway.
The agent market itself reshuffled just as much as the metrics did. ChatGPT’s share slid from 78% to 56% between July 2025 and July 2026, while Gemini climbed from 15% to 30% and Claude grew from 2% to 10%. Model choice has become a genuine business risk rather than a preference, which is one more variable that resists easy measurement.
Publishers Took The Fight To Courts And Regulators
Publishers spent H1 2026 fighting the same battle in a different arena. A Munich court ruled Google liable for false statements generated by AI Overviews. Four hundred newspapers sued OpenAI and Microsoft over unauthorized content use. The UK’s Competition and Markets Authority ordered Google to give publishers more control and transparency over how their content is used in AI search, including opt-outs from AI Overviews, AI Mode, and Discover summaries. Every one of those fights is really an argument over who gets to control access and attribution once a human is no longer the one clicking through.
The Second Half Splits Intelligence From Agency
Indig closes his report with a preview of the second half that deserves as much attention as anything in the H1 recap. He argues H2 2026 will “separate intelligence from agency.” Model capability keeps getting cheaper and more commoditized through open-weight competition, but the permission to actually act on someone’s behalf, spend money, reach data, or impersonate a person through an agent, is getting locked down harder by platforms, governments, payment networks, and users themselves. The publisher lawsuits and the UK CMA order are early proof that this tightening has already started.
That distinction matters more for SEO strategy than most of what got attention in H1. Here is what practitioners can do with it heading into the second half.
First, retire the assumption that a single rank-tracking tool covers AI search. Build a prompt panel that spans ChatGPT, Claude, AI Overviews, AI Mode, and at least one open-weight model, and treat the results the way a pollster treats a sample rather than the way an SEO treats a SERP. Given how little overlap Indig found between engines, tracking only one is close to tracking none.
Second, shift reporting away from citation counts and toward mention frequency, sentiment, and recommendation rank across that panel. A dashboard that only counts citations is measuring the smaller half of what actually drives buyer behavior, and it will keep understating your real AI footprint to clients or leadership.
Third, start auditing your agentic access layer now, well before H2 forces the issue. Check whether your product data, pricing, and checkout flow are structured so an authorized AI agent can actually complete a transaction on a customer’s behalf, and whether your brand shows up as a trusted, nameable option when an agent is choosing among several competitors. Indig’s split between intelligence and agency means the winners in H2 will be the brands agents are permitted to act on behalf of, not simply the brands that get mentioned most often.
I agree with Indig that the ability to measure AI’s impact fell behind the impact itself in H1 2026, and I do not think that gap closes on its own between now and the end of the year.
More Resources:
- Google’s AI Search Data Is Growing, But The Gaps Remain
- Google Announces AI Mode Checkout Protocol, Business Agent
- State Of AI Search Optimization 2026
Featured Image: Accogliente Design/Shutterstock
