The state of AI-driven search in 2026: an assessment for owner-managed businesses

The intent of this annual stocktake is to give an owner-managed business a frank, vendor-neutral view of where AI-driven search sits at the time of writing, what has materially changed in the last twelve months, and what the realistic implications are for the firm's website over the next eighteen months.

The piece is deliberately written for the buyer, not the practitioner. The technical detail is in the rest of this insights collection; this is the briefing.

The state of the search market

The most useful framing remains the one that emerged in late 2024: classical search and generative search now operate as parallel layers of the same market. A prospect researching a service provider is, in 2026, materially more likely than they were a year ago to ask an assistant rather than type a query into a search engine. The classical search interface hasn't gone away; the share of research behaviour that runs through it has, however, declined.

The implication for the firm is that the discoverability programme is now a portfolio. Classical SEO continues to matter, but the AI-search work (the GEO discipline) now sits alongside it as an equal investment.

The major generative search interfaces (ChatGPT search, Perplexity, Claude, Microsoft Copilot, Google's AI Overviews and Gemini) have converged on a similar approach to answer construction. The differences between them are now mostly about which sources they prefer, not about how they synthesise the answer. A firm optimising for one is largely optimising for all.

What changed in the last twelve months

Three substantive shifts.

Schema.org markup is now treated as a hard prerequisite. A site without proper structured data is materially harder to surface in any of the major generative interfaces than it was a year ago. The threshold appears to have risen. Where in 2024 a firm could rely on the model bridging gaps in the structured data using prose content, in 2026 the model is less forgiving.

llms.txt adoption has moved from "early adopter" to "common practice". The convention is now standard in domains where AI discoverability is taken seriously. The practical effect is that a firm without one is increasingly the exception, and the model's ability to construct a coherent summary of a firm in the absence of one has correspondingly declined.

The major template platforms have caught up partially. Wix, Squarespace and (to a lesser extent) Webflow have shipped meaningful improvements to their default structured-data output. The gap between a properly optimised template-platform site and a custom-built site has narrowed, although it hasn't closed. (See the vendor assessment for the specifics.)

What didn't change

Two things are worth highlighting.

The fundamentals are stable. Schema.org markup, llms.txt, FAQ schema, content depth, authority signals, recency, consistency: the six signals that mattered a year ago still matter. A firm that addressed them in 2024 hasn't had to redo the work in 2026. The investment is durable.

The economics of being missed haven't improved. A firm that isn't in the candidate set for the queries it ought to be the canonical answer to is missing the same volume of research traffic it was missing a year ago, only the underlying volume of research conducted through generative interfaces has grown. The opportunity cost of doing nothing has increased.

The realistic implications for an SMB

Three things follow.

First, the foundational work is now table stakes rather than competitive advantage. A firm that ships a website without proper structured data, a llms.txt, and FAQ markup is in 2026 below the median rather than at it. The correct framing is "do the foundational work to remain competitive", not "do the foundational work to gain an edge".

Second, the differentiation now sits in content depth and authority. The firms that are reliably cited by AI assistants in their categories are firms that have invested in sustained, substantive publishing on the topics they want to be associated with. The foundational work makes the firm legible to the model; the content depth makes the firm preferable.

Third, the cost of competent execution has fallen. The foundational work is, for a typical professional services SMB, a one-off engagement in the £2,500 to £5,000 range. The ongoing publishing cadence is in the order of two substantive pieces a month. Neither is a heroic investment relative to the firm's other marketing spend.

Where to start

For a firm that hasn't previously invested in this area, the right sequence is:

1. Audit. Establish where the firm stands against the six signals. The audit framework sets out the assessment; the Audit package productises the work.
2. Retrofit. Address the structural gaps surfaced in the audit: schema.org, llms.txt, FAQ markup, principal Person markup. The Retrofit package covers this.
3. Plan the content investment. Define the firm's topical positioning, the topics on which it wants to be the canonical source, and a sustained publishing cadence.
4. Execute. The Authority package bundles the build with the initial content investment. The maintenance retainer covers the ongoing publishing.

The implementation roadmap covers the sequencing in finer detail.

A note on what comes next

I'm asked, with some regularity, what the next shift in this market will be. The honest answer is that the fundamentals (being a credible source the model can read) are unlikely to change. The mechanism by which models identify and weight credibility will continue to evolve; the specifics of which signals carry more weight will shift; and new interfaces will continue to emerge. A firm whose foundational work is in place is, in my view, well-positioned to absorb those shifts without further structural investment.

The discipline is maturing. The work is becoming methodical. Owner-managers can engage with it as a finite project rather than an open-ended commitment.