01 / THE MECHANIC
A procurement lead at a mid-sized manufacturer needs a quality management system. Three years ago she opens eight tabs, reads two comparison articles, fills in a form, waits for a demo.
This year she types a paragraph into an assistant instead. Requirements, budget band, the compliance standards she has to satisfy, the ERP she needs it to talk to. Ninety seconds later she has a table with four vendors in it, assembled from public documentation, with pricing models and certification status already filled in.
You are not one of the four.
Your homepage is fine. Your case studies are good. Your sales team is better than two of the companies on that list. None of it was consulted, because none of it was legible to the thing doing the consulting.
02 / THIS IS NOT A FORECAST
MarTech described the shape of this in early August 2026: buyers moving off the old discovery model of keyword searches, review blogs and open tabs, and onto what it called automated delegation. The agent filters options against functional requirements, builds the comparison matrix, tracks pricing, and in the more advanced setups completes the purchase once conditions are met.
The mechanics are worth being precise about, because they are not a faster version of what came before. An agent does not interpret intent through keyword variations. It takes a natural-language instruction, scans indexed repositories, drops what does not fit, and returns a direct comparison. The search results page is skipped entirely.
Meanwhile the ground underneath ordinary search has been moving in the same direction. SparkToro's analysis of Similarweb clickstream data found that 68.01% of US Google searches ended without a click in the first four months of 2026, up from 60.45% in 2024. AI Overviews appeared on more than 20% of searches, and where they appeared, click-through fell by close to 60%.
I would not lean too hard on any single one of those numbers. SparkToro is candid that its historical figures come from different panels and providers, so the year-on-year comparisons are indicative rather than exact. The direction is what matters, and the direction has been consistent for six years.
03 / THE PART THAT SHOULD WORRY YOU MORE
AI Mode barely registered in that study window. Just 0.34% of searches moved into it between January and April. Then Google said at I/O 2026 that AI Mode had passed a billion monthly users and that query volume was more than doubling every quarter.
So the honest position is this: the share of B2B research currently mediated by agents is small, and it is compounding. Anyone quoting you a confident percentage is selling something. What you can say with reasonable confidence is that the curve is steep and the preparation is slow, which is an uncomfortable combination.
There is a second-order effect that gets missed. When an agent builds the shortlist, your competitor is not the vendor with the better product. It is the vendor whose product data was easier to parse. That is a strange thing to lose on, and most companies will lose on it without ever seeing the loss show up in a report.
04 / WHAT THE AGENT ACTUALLY READS
Almost every B2B digital presence was built for human navigation. Visual hierarchy that guides the eye. Benefit-led copy above the specifications. Technical detail behind a form, because that is where the lead capture lives.
Agents ignore all of it. They parse schemas, follow semantic markup, and weight their recommendations by how often and how authoritatively a brand is cited by independent third parties. MarTech put it plainly: in an agentic environment, market share is decided by technical data architecture.
Which means the gate is now a tax. We have spent twenty years teaching marketers to put the useful material behind a form, and that instinct has quietly acquired a cost. If your API documentation requires registration, or your specifications live inside an untagged PDF, an agent will either miss them or rank you below a competitor whose equivalent material is sitting in the open.
The same logic applies to pricing. A vendor who publishes a pricing model, even a range with conditions attached, can be placed in a comparison. A vendor whose only answer is "contact us" cannot be compared, so it gets left out of the comparison. Not penalised. Omitted.
05 / WHY THIS ISN'T AN SEO JOB
Here is where most companies will get the response wrong, and it is the reason we keep writing about sequencing.
The fix touches four things at once. Brand decides the sentence an agent uses to describe you, which means positioning has to survive being compressed into one line by a machine with no interest in your tone of voice. Marketing decides whether independent sources cite you at all. Digital experience decides whether your content is structured or merely pretty. Technology decides whether your specifications, pricing and documentation exist as data rather than as design.
Hand that to an SEO agency and you get schema markup on a site whose positioning is still vague. Hand it to a brand studio and you get sharper language buried in a PDF. Hand it to your web developer and you get clean markup describing the wrong thing. Each vendor does competent work. The result still fails, because the failure lives in the seams between them. We wrote about that cost in an earlier note, and agentic discovery is the sharpest version of it we have seen yet.
It also breaks your measurement. Click-through rates, impressions and keyword positions all assume the audience is a person with eyes. When the audience is a system deciding whether to include you in a list, those numbers measure the wrong thing, and budget calibrated to them quietly stops working.
06 / SIX QUESTIONS TO RUN THIS WEEK
None of this requires a platform migration. Start by finding out where you stand.
Ask three different assistants to recommend vendors in your category, using the words a buyer would use rather than your own category label. Note who appears and how you are described when you do. Open your highest-value product or service page. Can a machine extract what you do, who it is for, and what it costs, without inference? List everything useful currently sitting behind a form. Decide which items are genuinely lead magnets and which are just habit. Check whether your specifications, integrations and compliance certifications are in indexable HTML or trapped in PDFs and images. Find out who cites you that you do not own or pay for. If the answer is nobody, that is your largest gap, and it is the slowest one to close. Look at whether your analytics can separate agent traffic from human traffic at all. Most cannot.
Six answers will tell you more than another quarter of keyword reporting.
07 / THE COUNTERWEIGHT
I want to be careful not to hand you a panic.
Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, on cost, unclear value or weak controls. A separate signal points the same way: WordPress VIP's Future of the Web 2026 report found enterprise marketing teams spending 16.6 hours a week trying to influence how their brands appear in AI answers. That is most of a working day, every week, going into a problem that is mostly publishing mechanics wearing a strategy costume.
So the winning move is not to buy an agent. Plenty of companies are about to spend heavily on deploying their own and get very little back. The move is to be legible to everybody else's, which is cheaper, duller, and considerably more likely to work.
The funnel has not disappeared. It has moved earlier. Contracts still get negotiated by people, approvals still need signatures, and the relationship still decides the renewal. What changed is where human attention enters. The agent produces the shortlist, the human reviews it, and if you are not on it, the quality of your sales team is irrelevant. You are not being outsold. You are being filtered out before selling begins.
Sources
- MarTech, Optimizing your B2B brand for autonomous AI shoppers, August 2026
- SparkToro, In 2026, Less than One Third of Google Searches Still Send a Click, June 2026 (Similarweb clickstream panel, US, January to April 2026)
- Google I/O 2026, AI Mode usage figures
- Gartner, agentic AI project cancellation forecast
- WordPress VIP, Future of the Web 2026