AI halftime report: H1 2026

Each six months, I take a pointy take a look at the most recent developments in AI and Search. Primarily based on how shortly issues are growing, I’d virtually want to extend that cadence to month-to-month. (And possibly I’ll.) However first, right here’s what occurred in H1 2026.

H1 2026 moved cash, site visitors, jobs, and market cap earlier than anybody may show how a lot worth AI created. Search habits modified, token budgets exploded, software program shares bought off, and corporations blamed layoffs on AI.

Each main AI story in H1 2026 was actually a narrative about attribution.

• We’re nonetheless determining enhance accuracy in measuring AI visibility.
• Corporations spent billions on inference, however there isn’t a apparent reply to “What’s the ROI?”
• Public markets punished software program corporations, however had been buyers reacting to precise or perceived disruption?
• We’re listening to “AI brought about layoffs,” however while you look deeper, the actual causes aren’t linked.
• It’s clear the place the site visitors loss is coming from, nevertheless it’s not clear what content material market mannequin will change it.

The frequent thread: AI’s financial affect is increasing sooner than we will attribute it.

AI Search

Within the H1 2025 report, I predicted Google would roll AI Mode out additional, and that turned out to be true. AI Mode is now a click away from AI Overviews, which implies it’s simply two clicks away from common search outcomes.

To know the affect of this shift, I published a collection of person habits research and knowledge analyses within the first half of 2026.

The by means of line of Progress Memo analysis in H1 2026:

• Monitoring a model’s presence in AI Search is very advanced. You might want to issue within the engine, personalization, reasoning ranges, mannequin updates, stochastic variability, and so forth. Measurement is fragmented. The quotation/point out overlap between engines is minuscule: 91% of citations seem in solely one in every of ChatGPT, Perplexity, or AI Overviews. Immediate monitoring must be nearer to polling and focus teams than search engine optimization rank monitoring.
• Model mentions in AI solutions are extra impactful on enterprise outcomes than citations. Don’t get me flawed, citations form the reply, and for some companies, that’s most essential. However the majority of distributors and retailers want to concentrate to how often they present up in a panel of prompts, by which context, what sentiment, and whether or not they’re advisable or not (forward of rivals).
• Belief is the best foreign money in AI search. Near 75% of shoppers choose the primary end in an AI shortlist. But when they see a trusted model anyplace on that record, they may choose it.
• The common U.S. grownup has excessive belief in AI suggestions. 88% of the time, customers accepted AI Mode product suggestions as greatest there’s. In AI Overviews, alternatively, users click, evaluate, and compare so much. That’s traditional search habits, nevertheless it doesn’t carry over to AI chatbots.
• Writing model, token price range, and technical elements matter for brokers. If you wish to be current in AI Search and agentic suggestions, it’s essential lead with unique information, write in a direct, easy-to-understand way, take away fluff and hold your web site technically quick and straightforward to entry.

AEO/GEO is a model channel disguised as a efficiency channel. AI suggestions form demand (not citations). The unit of optimization is whether or not AI names, trusts, and recommends your model.

The grAIpes are bitter

November 2025 was a pivotal turning level in AI. Claude’s Opus 4.5 was the primary mannequin that was perceived as dependable and strong sufficient for agentic workflows. A month later, Peter Steinberger went viral with Clawdbot, which led to millions of installs, long lines in China, and Nvidia constructing a Nemoclaw clone.

Then, issues bought even crazier. Shopify, Uber, Meta, and lots of different tech giants constructed token management boards that incentivized engineers to gentle as many tokens on fireplace as doable. 

“Let me offer you a thought experiment. Let’s say you’ve a software program engineer or AI researcher, and also you pay them $500,000 a yr. On the finish of the yr, I’m going to ask him how a lot did you spend in tokens. And [if] that individual stated $5,000, I’ll go ape one thing else. If that $500,000 engineer didn’t devour at the least $250,000 value of tokens, I’m going to be deeply alarmed.”

— Jensen Huang

A Meta engineer created an inner leaderboard dubbed “Claudeonomics” that ranked over 85,000 workers by the variety of AI tokens they burned. It led to large waste as workers left AI brokers working on idle or ineffective duties simply to climb the ranks and earn titles like “Token Legend”.

Reportedly, Meta engineers consumed 73.7 trillion tokens in simply 30 days. (I’m wondering how a lot of OpenAI’s and Anthropic’s development was ignited simply by workslop.)

Good instances. However in April, CFOs pumped the brakes after discovering that workers had burned by means of their annual token price range in 4 months. TokenMAXXXING shifted to valueMAXXXING. 

Meta’s leaderboards had been shut down in April (together with leaderboards at different corporations), however not earlier than the #1 person on the board averaged 281 billion tokens, which might value over $1.4 million at commonplace API charges.

However what stays from the all-you-can-eat token craze is a step change in productiveness that we can’t but measure. As I wrote, AI changed my work. And yours, too:

I’ve used AI to lift my affect by a magnitude.

• I designed touchdown pages end-to-end for a serious journey model that made it into manufacturing.
• I automated subject prioritization, search engine optimization testing, and search engine optimization reporting for my purchasers with full-blown apps.
• I constructed an array of helpful purposes for myself, from automating the SSI (search engine optimization Website Index discovered within the bimonthly Growth Intelligence Briefs) to Openclaw brokers that assist me with analysis and charts.

One consequence of the token tsunami is that much more folks began utilizing Claude, which in return fired up the expansion engines for AI infrastructure corporations that Claude advisable, like Supabase.

SaaS massacre

In February 2026, as a consequence of the Opus 4.6 launch and Claude Cowork, the software-as-a-service business noticed a extreme market downturn that worn out tons of of billions of {dollars} in market cap over just some buying and selling classes. The SaaS vertical noticed a steep annual decline of +30% within the markets. Valuations for high-growth software program corporations collapsed, with income multiples shrinking dramatically in comparison with their pandemic highs.

In case you’re working in B2B, you’ve in all probability felt the affect within the type of decrease conversion charges, longer gross sales cycles, and softening model searches.

The Brand Tax Calculator offers you the greenback determine for the hole between branded paid spend and reported ROAS, and the AI SEO Budget Reallocation Planner turns it into 5 growth-priority eventualities in beneath 7 minutes. Each stay within the Premium Resource Library. 

It seems not all software program was (or is) forsaken. It’s the underside quartile of software program corporations that pulls the market down. Curiously, the decline is solely related to how “the market” views an organization’s AI robustness. Firm valuation drops usually are not associated to efficiency, however reasonably as to if folks suppose AI will disrupt their product.

During the last 30 buying and selling days, each the highest quartile and median IGV inventory have outperformed the ETF as an entire — it’s the underside quartile (which embrace a number of the largest corporations) that’s knocking down total efficiency.

So, should you’re in B2B and both compelled or privileged to take a brand new job, choose an organization the market is optimistic about.

AI washing within the labor market

You may need learn or heard this just lately: “We’re shedding x% of our workforce attributable to AI.” It’s the narrative du jour. Challenger, Gray & Christmas reported that AI was the main cited purpose for Might 2026 job cuts and had been cited in 87,714 cuts year-to-date, equal to 22% of all 2026 introduced layoffs by means of Might.

And don’t get me flawed. Layoffs are actual:

• Tech layoffs up ~66% yr over yr, towards 150K
• Oracle lower 21,000 citing AI
• Challenger lists AI as the highest cited purpose (~87,700 YTD by Might)
• Block lower 40% and its inventory rose (Bloomberg’s “AI washing” framing)

The one drawback is that AI isn’t the trigger. I’ve written extensively concerning the Huge Labs utilizing AI substitute as a advertising narrative. Nevertheless it hasn’t come true but. A whole lot of AI layoffs had been actually to offset capital expenditures, pandemic overhiring, and financial turbulence. We are able to all think about a future by which AI is so good that it makes sure roles out of date, however that point isn’t now.

A yr in the past, within the H1 2025 report, I predicted that AI layoffs had been a PR stunt. And I used to be proper: Regardless of clear productiveness positive aspects from AI, the current waves of layoffs present no indication of AI being the underlying trigger. In actual fact, corporations utilizing AI as a purpose to put workers off are rehiring them.

The agent market fragmented

ChatGPT’s market share decreased from 78% in July 2025 to 56% in July 2026, whereas Gemini went from 15% to 30% and Claude from 2% to 10%.

Google is now the almost definitely participant to win the AI shopper market, and ChatGPT has misplaced its significance in AI Search to Google:

• ChatGPT nonetheless has ~1.1B customers, however OpenAI refocused on an enterprise pivot and shut down Sora, dropped video in ChatGPT and killed On the spot Checkout. Enterprise makes up ~40% of revenue, heading to ~50% forward of a possible IPO. I’m skeptical that we’ll be capable of purchase OpenAI inventory. In line with Ed Zitron, OpenAI spends $2.8 billion USD each month to make $1.1B.
• Anthropic’s battle with the Pentagon pushed Claude to No. 1 amongst free apps on Apple’s U.S. App Retailer. Just a few months later, customers appear much less positive if Anthropic can resist assaults from open-source fashions.
• The Huge Labs massively subsidize tokens. SemiAnalysis discovered a $200 USD plan of the frontier fashions gives you 8-14K USD worth in tokens. 

These days, open-source fashions like Kimi K3, GLM 5.2, or Deepseek V4 have put stress on the large US labs due to these subsidies. Open-weight and open-source fashions additionally present that the mannequin itself turns into commoditized, whereas the harness and utility layer achieve extra significance. 

Ahead-deployed-engineer job postings rose 800% within the first 9 months of 2025 and had been nonetheless up 729% year over year in April 2026. Frontier Labs now pays more than $500,000 in total compensation for individuals who can flip a mannequin demo right into a working system inside a buyer’s enterprise.

By no means have entrepreneurs confronted a sooner evolving surroundings with extra platforms to measure… which is why understanding what platforms matter most in your viewers is essential. However from an execution perspective, mannequin availability has develop into a enterprise danger, as we will see on the hand of Fable 5. Having choices is nice.

Publishers and the regulation

I predicted as much as 70% of 2024 natural site visitors could possibly be passed by 2026, however that was overeager. In the long run, the numbers land at ~33% referral loss year over year, with 68% of Google searches now ending without a click. Nonetheless, in consequence, publishers are transferring the battle to the courts and regulators.

USA Immediately CEO Mike Reed explained: “We’re getting shut now to the purpose the place we’ll block Google as effectively and abandon the standard search site visitors that we get as we speak.” 

Listed here are just some lawsuit-related highlights:

In response to the extraordinary publishing panorama adjustments, Parallel AI and Cloudflare are each making an attempt to construct publishing marketplaces for content material house owners and AI corporations. 

“Assume there’s no search. It’s important to have your companies deliberate as if search is zero.”

— Conde Nast CEO Roger Lynch

Publishers try to exchange the content-for-traffic discount with a content-for-training market. This leaves a number of open questions:

• Who units the value and the way? 
• How will you show that an article contributed to an AI agent’s reply or buy? 
• Do publishers have negotiation energy? 
• Are smaller publishers even wanted by AI labs?

Dig in: Full record of Progress Memo analysis accomplished in H1 2026

Dig deeper into the patterns coated on this halftime report beneath:

• Person habits research:
• Information analyses:

As we go the midway level within the yr, a giant because of my editorial and analysis companions, Amanda Johnson and Eric van Buskirk. And an extension of gratitude to my knowledge companions who make these analyses doable: Semrush, AirOps, Gauge, Quotation Labs, Siteline, SalaryGuide, and AudienceKey. 

This publish first appeared on the writer’s web site and is republished right here with permission.