Intelligence Brief · Web Intelligence · April 2026

Search Has a
New Front Door.

By Frank Graham

AI is intercepting your highest-intent customers before they ever reach your website. This brief explains what is happening, why it matters for your specific sector, and where your business stands in the new economy.

1.8bn

UK ChatGPT visits in 8 months of 2025, up from 368m the year before

Ofcom Online Nation, December 2025

−58%

Average click-loss on the top organic result when Google shows an AI Overview

Ahrefs, December 2025 · ~300k keywords

8% vs 15%

Click-through when an AI summary appears in Google’s results, versus when it does not; nearly half

Pew Research Center, July 2025 · 68,879 queries

“The buyer’s first question is no longer ‘which result is best.’ It is ‘which answer am I being given.’”

I The Shift

The decade-old playbook has expired.#

For ten years, the formula was simple. Rank on page one. Generate clicks. Work your funnel. That formula still holds for a shrinking slice of queries. For the rest, AI has already answered the question before your listing appears.

54% of UK adults now use AI tools; rising to 79% of 16–24-year-olds and 74% of 25–34-year-olds (Ofcom, Adults’ Media Use and Attitudes, April 2026). 94% of B2B buyers report using ChatGPT, Copilot, Perplexity, or Claude during the purchase journey (Forrester / 6sense, 2025). Your buyers are not opening five tabs to compare options. They are asking AI for the final answer.

UK AI Overviews launched on 15 August 2024. We have now been operating with AI-summarised search results for the better part of two years. The shift is not approaching. It has happened, and it is now compounding.

The shorthand for what comes next is the Agentic Web: a web in which AI agents (not users sifting through ten blue links) synthesise information, weigh sources, and make active recommendations on the buyer’s behalf. The unit of competition is no longer the click. It is the citation.

If your brand is not part of the AI answer, you do not exist in that buyer’s world. Not because your product is inferior. Because your website is not structured to be read, understood, and cited by the systems making the recommendation.

There is a useful frame for what has changed. For a decade, every serious website optimised for User Experience: speed, clarity, conversion, delight. UX was the discipline that won. The next decade will be won on Agent Experience: speed of comprehension, clarity of fact, extractability of structure, density of citation. AX is what the new front door is built for. The brands that recognise this early will compound. The brands that do not will be invisible by the time they notice.

This brief explains the mechanics, and gives you a framework to diagnose where your business stands today.

“The Agentic Web does not grade on a curve. It either recommends you, or it does not.”

II The Maturity Curve

Every business on the web sits on one of four rungs.#

The Agentic Web does not grade on a curve. Each transition between rungs requires a specific, identifiable change, and the changes are not evenly distributed in cost or complexity. Most businesses are stuck not because the fixes are expensive, but because nobody has told them which fix applies to them.

Indexing is not citation. Citation is not recommendation. Most playbooks confuse the first with the last.

01 · Rung One

Invisible

AI crawlers find nothing credible to work with. No structured data, no citeable facts, possibly blocked from crawling entirely. When a buyer asks about your category, you are absent from the answer.

Diagnosis. Search your own brand in ChatGPT. If the model cannot describe you correctly, cite a source, and identify what you sell and to whom; you are here, regardless of what your SEO dashboard says.

The move to Emerging

Unlock the front door.

The most common cause of Invisible status is a robots.txt file blocking AI crawlers, often set years ago during a GDPR or security hardening exercise and never revisited. GPTBot, ClaudeBot, and PerplexityBot must be explicitly permitted.

Audit robots.txt; permit GPTBot, ClaudeBot, PerplexityBot, GoogleOther
Add one Organisation or LocalBusiness schema block
Ensure core facts (name, location, what you do, who you serve) exist as plain HTML; not inside JavaScript, images, or iframes

Effort: a single sprint. Zero budget beyond developer time.

02 · Rung Two

Emerging

Partial signals exist. Some schema, some third-party mentions, but fragmented. AI occasionally surfaces your name but cannot accurately describe what you do or why you are different.

Diagnosis. Ask the model your category question; “Recommend a [your category] in [your geography].” You may be mentioned, but you are not in the first three names. The citation graph is recommending someone else, and that someone else is winning the enquiry you used to win.

The move to Cited

Give AI something worth citing.

Emerging businesses are visible but unreliable. The transition to Cited requires extractable, attributed, specific content; not more content, better-structured content. A peer-reviewed Princeton study at KDD 2024 (Aggarwal et al., Generative Engine Optimization) measured this directly: the top-performing on-page methods (adding verified statistics, incorporating attributed quotations, and citing credible sources) delivered a 30–40% relative improvement in citation visibility, with the strongest single result a 41% lift on Position-Adjusted Word Count.

FAQ schema on your most-queried pages; the highest-leverage single schema type for citation frequency, shown to double citation frequency in independent A/B testing (Growth Marshal, Perplexity Sonar analysis, June 2025)
A named expert or founder bio with verifiable credentials, as plain text on the page
At least one specific, attributed statistic per key service page
Pricing or service scope visible as HTML text; not hidden behind “contact us”

Effort: two to four weeks. Content-led, no significant rebuild required.

03 · Rung Three

Cited

AI reliably mentions you. Your core facts are extractable. But you are cited passively; appearing when queried, not yet positioned as the recommended answer.

Diagnosis. Ask the model the buyer’s real query; “Who should I hire for X?” Not “What is X?”; that is a researcher’s query. The hire query separates Cited from Recommended. You only sit at Recommended if your name is in the first sentence of the answer.

The move to Recommended

Become the answer, not just a source.

Cited businesses appear when queried. Recommended businesses are selected. AI leads with their name, not alongside three competitors. The transition requires authority signals that extend beyond your own website.

H1s and opening paragraphs structured to answer likely AI queries directly; the first sentence is the citation
Active presence on the three to five platforms your sector’s AI trust hierarchy treats as authoritative
Attributed client quotations on the page; the same KDD study found attributed voice among the strongest single citation drivers in generative responses
Consistent brand signals across every AI-indexed property: your own site, third-party profiles, trade press mentions

Effort: a 60–90 day programme. The rung most businesses cannot reach without an audit first.

04 · Rung Four

Run this in five minutes.

“You solved for User Experience for ten years. The next ten will be won on Agent Experience.”

III The Mechanics of Citation

You solved for User Experience.
Now you must solve for Agent Experience.#

You have spent years making your website beautiful for humans. Human-beautiful websites are often invisible to AI. The mechanics of AI citation are documented, and they are not what most marketers expect.

AI crawlers operate on a token budget. If your site is bloated, renders late, or buries core facts inside JavaScript, the AI burns through its budget and leaves without understanding what you do. Your competitor (with a technically inferior product but a cleaner information architecture) gets cited instead.

The Princeton/KDD work on generative engines (Aggarwal et al., 2024) provides the most rigorous public evidence of what actually moves the needle. Three findings stand out for any business being read by an LLM.

Lever	Finding	Source
Statistics & verified data	Adding specific, attributed numbers was among the top three methods, contributing to the 30–40% relative improvement in citation visibility delivered by the best on-page strategies.	Aggarwal et al., KDD 2024
Direct attributed quotations	Adding credible, attributed quotations was the single most effective method in the People & Society, Explanation, and History domains in the study.	Aggarwal et al., KDD 2024
Structured data (JSON-LD)	Independent A/B testing on Perplexity’s Sonar found pages with FAQ schema doubled citation frequency and were cited roughly six hours faster than non-schema controls.	Growth Marshal, June 2025

There is no single “AI visibility.” There is ChatGPT visibility, Perplexity visibility, Gemini visibility, and they can diverge sharply by vertical.

A URL cited reliably in one engine may be invisible in another. Monitoring requires model specificity, not a single score. The brand that scores well across three or more engines is the brand the citation graph has converged on. The brand that scores in only one is winning a fluke, not a seat.

IV The Failure Modes

Two failure modes nobody is pricing in.#

Beyond the visibility curve, two specific failure modes are costing businesses enquiries and revenue silently, before a buyer has ever visited the website. These are the mechanics most marketing leaders have not yet absorbed.

The Vendor Bake-Off Problem

Buyers are asking AI to compare vendors directly; pricing, features, support, implementation time. The query is no longer “what are my options”; it is “compare Vendor A and Vendor B on price, features, and customer support.”

If your pricing is hidden behind “contact us”, buried in a PDF, or locked inside an image, AI cannot include you in the comparison. It will either hallucinate your features or skip you entirely, and recommend a competitor with a simpler information architecture.

You lose the bake-off silently, before the buyer has ever visited your site. You will never know it happened. There is no analytics event for “AI declined to mention you.”

The Citation Consensus Problem

AI does not only read your website. It reads everything the internet has already said about you; niche forums, sector publications, trade press, third-party review sites, regulator listings. For every vertical, there are specific platforms AI treats as authoritative.

For consumer queries: Reddit, increasingly. For professional services: specialist trade press, industry bodies, professional registers. For hospitality: sector-specific review aggregators, editorial guides, and named-critic coverage.

Being absent from the platforms your sector’s AI trust hierarchy uses is as damaging as having a broken website. Most businesses have no idea which platforms matter for their specific vertical; which means they are competing against rivals who have already cornered the citation graph without anyone noticing.

This is why monitoring across three engines matters. It is not a thoroughness exercise. It is the only way to see whose citation graph the models are converging on. The answer is not always who you would expect.

V Evidence From Your Sector

This is not a distant trend.
It is live in your market, right now.#

The shift looks different by vertical, but it is happening across all of them, and in several sectors the infrastructure for AI-intermediated commerce is already in place.

Hotels, Dining & Pubs

Guests are choosing where to stay, eat, and drink inside AI conversations.

One in three travellers now use AI to plan or book trips, up from a much smaller base in 2024 (Phocuswright, August 2025; Adyen, 2025 Hospitality and Travel Report). Google AI Overviews intercept the discovery searches that fill dining rooms and function spaces; “pubs near me with a garden”, “best Sunday lunch in [city]”; displaying hours, menus, and booking links without a single click to your site.

Signal: ChatGPT outbound referral traffic to the open web grew 206% year-on-year in 2025, and the number of unique domains receiving ChatGPT referrals tripled from roughly 71,000 to 260,000 over the same period (Semrush, February 2026). The discovery layer is widening, fast, and the brands inside it now will compound the advantage.

B2B & Professional Services

Google is now the validation step, not the starting point.

94% of B2B buyers use ChatGPT, Copilot, Perplexity, or Claude during the purchase journey (Forrester / 6sense, 2025). Research happens in AI. Shortlisting happens in AI. Google is used afterwards to confirm what AI has already decided. For accountancy, legal, and consultancy, the discovery conversation has already moved.

Signal: Gartner forecasts that by 2028, 90% of B2B buying will be AI-agent intermediated, with over $15 trillion of B2B spend flowing through AI-agent exchanges (Gartner press release, 21 October 2025). That is not a directional prediction. That is a category-level forecast from the firm enterprises use to set IT strategy.

Private Members’ Clubs

Prospective members are forming a view of your club before they ever apply.

At the senior-professional level the club exists to attract, AI assistants are increasingly the first point of comparison. Asked about the category, ChatGPT and Claude describe each option’s atmosphere, cuisine, and member profile from whatever the wider web has said about it; press coverage, member directories, review aggregators. Without structured information about location, programming, and membership terms exposed in a way AI can extract, the model’s description of the club is whatever editorial happens to capture, not what the club’s own site says.

Signal: The same B2B research that puts 94% of professional buyers using AI assistants during the purchase journey (Forrester / 6sense, 2025) describes the demographic that joins clubs. The decision moment is moving inside the same tools they use for work.

Venues & Events

Couples are shortlisting venues without ever opening a browser tab.

36% of couples now use AI tools for wedding planning; nearly double the rate at the start of 2025 (The Knot Worldwide, February 2026). And the infrastructure for AI-mediated booking is already live: in February 2026, The Knot Worldwide launched the wedding industry’s first dedicated app inside ChatGPT, surfacing its vendor marketplace (venues, photographers, caterers) directly within the chat interface (BusinessWire, 3 February 2026).

Signal: If your venue is not in that conversation, it is not on the shortlist. The discovery moment has moved.

VI The Timing Question

The question is when the tipping point reaches your vertical.#

AI Overviews are already eroding click-through on the queries that trigger them; by 58% on the top organic result (Ahrefs, December 2025), with the click rate falling from 15% to just 8% on AI-summarised pages (Pew Research, July 2025).

The businesses being cited in AI answers today are building a compounding advantage. Citation positions reinforce themselves: models retrain on a web that already cites you, third-party platforms reference you because the models do, and the citation graph thickens around your name. The brands that are Recommended in AI results 18 months from now are the ones being audited today.

The question is not whether to act. It is whether you have the lead time to act before your vertical tips.

Where You Stand, Specifically

You cannot fix what you cannot see.#

If you scored 0, 1, or 2 on the audit; that is the gap a Web Intelligence Report closes. The rungs are universal. The fixes are not. Which robots.txt lines, which schema, which third-party platforms, which competitors are eating your citation share; those answers are specific to your site, your sector, your buyers.

Commission a Web Intelligence Report → £1,000 · Delivered within 48 hours · View all pricing options Not ready for the full report? Start with AI visibility only; £250 →

The Cost of Not Knowing

The first compounding curve in digital visibility since PageRank. Why most agencies sell upside, and what happens when you quantify downside instead; with a worked methodology and three case studies from Q1 2026.

Continue reading →

The Shift Briefing

Citation behaviour shifts as models retrain.

The rules of Generative Engine Optimisation are being written in real time. Retrieval patterns shift. Citation factors evolve. We track it monthly and share what matters for UK businesses, no consultancy noise, just the intelligence.

Subscribe to The Shift Briefing No pitch sequence. Unsubscribe any time.

Sources & references

Ofcom, Online Nation 2025: published 10 December 2025. ofcom.org.uk
Ofcom, Adults’ Media Use and Attitudes Report: published 2 April 2026. Source for 54% / 79% / 74% AI tool adoption.
Ahrefs, AI Overviews Reduce Clicks by 58%; Update: Ryan Law, December 2025. 300,000-keyword sample. ahrefs.com
Pew Research Center, Google users are less likely to click on links when an AI summary appears: July 2025. 68,879 queries from 900 US adults. pewresearch.org
Forrester / 6sense, 2025 Buyer Experience Report: survey of ~4,000 global B2B buyers; 94% LLM-usage finding.
Gartner, Top Strategic Predictions for 2026 and Beyond: press release, 21 October 2025. “By 2028, 90% of B2B buying will be AI agent intermediated, pushing over $15 trillion of B2B spend through AI agent exchanges.” gartner.com
Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., Deshpande, A. GEO: Generative Engine Optimization. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD ’24), Barcelona, August 2024. arXiv:2311.09735. arxiv.org/abs/2311.09735
Growth Marshal, The 2025 Perplexity Playbook: Sonar Ranking Factors: Kurt Fischman, June 2025. Source for the FAQ-schema A/B testing results and structured-data ranking.
Phocuswright, AI in Travel Planning; 2025 Update: August 2025. Source for one-third of US travellers using AI for trip planning.
Adyen, 2025 Hospitality and Travel Report: August 2025. Source for one in three global travellers booking with AI, up 74% YoY.
Semrush, ChatGPT Traffic Analysis: Insights from 17 Months of Clickstream Data: February 2026. Source for 206% YoY growth in ChatGPT outbound referrals and the 71k → 260k unique-domain expansion. semrush.com
The Knot Worldwide, Launches the Wedding Industry’s First App Within ChatGPT: BusinessWire press release, 3 February 2026. Source for the 36% AI-tool adoption figure and the integration launch. businesswire.com

Search Has aNew Front Door.