The Answer Engine: How the Click Economy Ends and What Agencies Publish Instead
Your reporting was built for a channel that returned clicks. That channel is closing. This is what replaces the numbers on the monthly deck - and what has to get written to earn them.
For three decades, search worked as a simple transaction: a user typed a query, received a ranked list, and clicked through to evaluate sources. Every business competed for position on that list. The list produced clicks. Clicks produced revenue, and - not incidentally - produced the numbers agencies put on monthly reports.
That transaction is breaking down. In May 2025, Similarweb documented zero-click searches at 69% of all Google queries, up from 56% before AI Overviews launched a year earlier. Meanwhile AI-sourced referral traffic grew 527% year over year between January and May 2025 (Previsible, across 19 GA4 properties). Volume in the new channel is rising sharply; the share of search that produces a click to anybody’s website is falling just as sharply.
For an agency this arrives first as a reporting problem. The sessions line on the client deck goes down, the work has not got worse, and the explanation you have available is not yet a strategy. This page is about the strategy.
| Metric | Data point | Source |
|---|---|---|
| Zero-click search rate, May 2025 | 69% of Google queries | Similarweb, July 2025 |
| Zero-click rate before AI Overviews (May 2024) | 56% | Similarweb |
| Zero-click rate where AI Overviews appear | 83% | Search Engine Land, 2025 |
| AI-sourced traffic growth, Jan–May 2025 | +527% YoY | Previsible, 2025 |
| Gartner forecast: search volume decline by 2026 | −25% | Gartner, February 2024 |
| AI search conversion vs. Google organic | 14.2% vs. 2.8% | Superprompt, 12M visits, 2025 |
| ChatGPT weekly users, October 2025 | 800 million | Sam Altman via TechCrunch |
What the recommendation economy actually is
The recommendation economy is the environment where AI systems, rather than ranked lists, determine which vendors a buyer discovers - delivering one answer with implicit endorsement and no visible alternatives.
A traditional results page shows ten organic results, paid placements, and features competing for attention. A client with budget can buy their way onto the same visible field as a better-optimised rival. A recommendation is singular. When a buyer asks an assistant “which vendors handle contract lifecycle management for a mid-market legal team”, the assistant selects one answer, or at most a short list. No ads. No competing listings alongside. The filtering, evaluation and choosing already happened.
This changes what a client is buying from you. In the click economy they bought exposure and you reported on it. In the recommendation economy they are buying eligibility to be named, which is a harder thing to sell and a more durable thing to own.
The zero-click timeline, briefly
Zero-click search grew from a fringe concern to the dominant outcome through deliberate Google product decisions, then accelerated with AI Overviews.
Before May 2024, about 56% of Google searches ended without a click - featured snippets, knowledge panels and “People Also Ask” had already been chipping away for years. AI Overviews launched in May 2024. By May 2025 the zero-click rate hit 69%, and queries triggering an AI Overview showed 83%.
The downstream effect on publishers is the clearest available evidence of scale. Organic CTR on AI Overview queries dropped 61%, from 1.76% to 0.61% (Search Engine Land, 2025). News publishers lost more than 600 million monthly organic visits between mid-2024 and May 2025. HubSpot reported declines of 70–80% in specific categories. The median publisher was down about 10% year over year in the first half of 2025.
If your clients publish in order to earn sessions, that is the curve they are on, and no amount of craft reverses it within the old model.
Why the traffic AI does send behaves differently
AI-referred visitors convert at markedly higher rates than organic search visitors, because they arrive having completed most of their research inside the assistant rather than before starting it.
A Google user types a query, scans, clicks, reads, returns, clicks again, and eventually forms a view. They land on a site near the start of a decision. An AI search user asks a complete question, receives a synthesised recommendation, and - if they click at all - arrives specifically because the assistant cited that source as credible. The visit is a confirmation step, not a research step.
A 2025 Superprompt analysis of 12 million visits across 350 businesses found AI search traffic converting at 14.2% against Google organic’s 2.8%. Semrush’s June 2025 research found AI referral sessions converting at 4.4x organic. Passionfruit’s ecommerce data showed 11.4% against 5.3%.
Label these as market benchmarks when you put them in a deck. They describe a category, not a promise, and client-specific conversion belongs in the client’s own analytics rather than in a vendor’s claim.
The strategic reading is what matters: volume stops being the primary variable. Position in the answer takes its place.
What AI platforms do with client content
AI systems crawl aggressively to train models and build knowledge bases, and return a small fraction of that value as referral traffic. Cloudflare documented this throughout 2025 with the crawl-to-refer ratio - pages crawled per visitor sent back.
In July 2025, Cloudflare found Anthropic’s ClaudeBot at roughly 38,000 crawls per referred visitor, down from 500,000:1 in January 2025. OpenAI’s GPTBot spiked to 3,700:1. Perplexity, the most balanced, stabilised below 200:1 by late 2025.
The purpose breakdown compounds it. In July 2025, 79% of AI crawler activity was classified as training, 17% as search, and user-driven actions that could generate referrals represented 3.2%.
The implication for an agency is not to block anything. It is that a client’s content is already being read, processed and synthesised whether or not anyone is managing it. The only open question is whether the synthesis includes their name.
The KPIs that replace the old ones
Rankings, impressions, clicks and CTR measure exposure inside a shrinking click economy. Reporting in this channel needs different instruments, and putting them on the deck early is how an agency gets ahead of the awkward conversation rather than behind it.
Presence in answers. The primary competitive metric: across a defined set of buying questions in the client’s category, how often is the client named, and who is named instead. This is the number that replaces rankings, and it is the one clients understand immediately because it is phrased as a question a customer would ask.
Citation rate. Which specific pages assistants draw from, and how often. A page earning no citations is not being read, regardless of its traffic.
Recommendation frequency. Distinct from mere mention. Being cited as a source is not the same as being explicitly endorsed. This tracks the decision-stage prompts - the ones closest to a purchase.
Sentiment and framing. How assistants describe the client when they do mention them. Users tend to accept the framing they are given, so the adjectives matter commercially.
What the published work earns. Readership on the pieces you publish, click-throughs from them to the client’s own site, and when the assistants last came back and fetched them. This is the most concrete of the new metrics because it is measured on surfaces you control rather than inferred from someone else’s platform.
| KPI | What it measures |
|---|---|
| Presence in answers | Share of target buying questions where the client is named |
| Citation rate | Which pages assistants actually draw from |
| Recommendation frequency | Explicit endorsement, not just mention |
| Sentiment and framing | How the client is characterised when named |
| Published-work performance | Readers, click-throughs to the client site, last assistant fetch |
Why a recommendation is worth more than a click
A single recommendation to a high-intent buyer, in an environment with no ads and no visible alternatives, carries higher per-impression value than a search result - even at equal volume.
Consider what surrounds a search result: paid placements above, competitors adjacent, features answering the question without a click, and a back button one tap away. The environment is adversarial and attention is fragmented.
An AI recommendation works differently. The buyer asks in natural language, the assistant synthesises and presents an answer, and the user experiences it as disinterested curation. Consumer behaviour research in the Journal of Electronic Commerce (MDPI, 2025) found people perceive AI recommendations as more objective than human editorial recommendations, precisely because the assistant appears to have no commercial interest.
That perception is not strictly accurate - assistants are shaped by training data, content accessibility and entity prominence - but it is the user’s experience, and it drives behaviour.
What this does to the role of a client’s website
In the click economy, a website was the destination. In the recommendation economy it becomes the evidence an assistant reads before deciding whether to name the client. The visit may never happen.
This is a genuine repositioning, and it is worth walking a client through slowly. A business with excellent products, strong references and a beautifully designed site that blocks AI crawlers, hides content behind JavaScript, or publishes without extractable claims will not appear in answers regardless of how strong it is commercially. Assistants cannot recommend what they cannot read.
Three structural barriers come up repeatedly in client estates:
- Illegibility - key information embedded in images, PDFs, gated downloads or client-side-rendered components that crawlers cannot reliably reach.
- Entity ambiguity - inconsistent names, descriptions and facts across the web, making it hard for assistants to assemble a coherent picture of who the business is.
- No direct-answer content - pages that establish expertise at length but never state a plain, extractable claim an assistant can lift into a response.
The third is the most common and the least technical. It is also the one that requires actual writing rather than configuration, which is why it tends to be the last one addressed.
Building the reporting an agency can defend
Establish a baseline first. Before commissioning anything, measure how a defined set of buying questions is currently answered. Document which produce the client’s name, which produce only competitors, and which produce nothing recognisable in the category. That baseline is what every later report is measured against, and gathering it takes a fraction of the time clients expect.
Model the value per session with the client’s own numbers. Take their conversion rate and average deal value rather than a benchmark. A planning model built from their data survives scrutiny; one built from a vendor’s case study does not.
Track the attribution gap explicitly. Most analytics platforms currently misattribute a large share of AI-influenced traffic as direct. A buyer who receives a recommendation, closes the session, and navigates to the site later is recorded as a direct visitor. AI’s commercial influence is therefore systematically under-reported in standard dashboards. Say so in the report rather than letting the client discover it.
Report on what the published work did. Readership, click-throughs to the client’s site, and assistant fetch activity are directly observable on the pages you publish. They are narrower than a full attribution story and considerably more honest, and they are the material that should shape the next month’s brief.
What actually gets commissioned
The whole argument reduces to one operational question: what do you write next, for which client.
Answering it by intuition is expensive and hard to defend on a renewal call. Answering it by evidence means running a fixed set of buying questions across the assistants, seeing which ones return somebody else’s name, and treating that ranked list as the month’s brief. Then publishing against it, re-asking, and looking at what the published pieces earned in readers and click-throughs - and letting both of those signals steer the following month.
Two inputs, one decision. It is a working process with a named editor’s sign-off in it - yours or your client’s - not an engine that reassigns itself. But it turns a content plan from a proposal into something with evidence behind it, which is the part clients renew.
Frequently asked questions
What is an answer engine? A system that synthesises information and delivers one direct response instead of a list of links. Answer engines make the selection decision on the user’s behalf; search engines delegate it back to the user.
What share of Google searches end without a click? 69% as of May 2025, up from 56% before AI Overviews launched (Similarweb, July 2025). Queries that trigger an AI Overview reach 83%.
How fast is AI search traffic growing? AI-sourced traffic grew 527% year over year between January and May 2025 across 19 analysed GA4 properties (Previsible, 2025).
Why do AI-referred visitors convert better? They arrive after completing their research inside the assistant, already holding a recommendation. A 2025 Superprompt analysis of 12 million visits found 14.2% against Google organic’s 2.8% - a market benchmark rather than a guaranteed outcome.
What is the crawl-to-referral imbalance? The gap between how aggressively AI bots harvest content and how little traffic they return. Cloudflare recorded ClaudeBot at ratios as extreme as 500,000:1 in January 2025, improving through the year; Perplexity remained the most balanced below 200:1.
Does being cited in an AI Overview help or hurt clicks? Both, depending on position. Pages cited within an AI Overview earn about 35% more organic clicks than equivalent uncited pages, while queries triggering an AI Overview reduce organic CTR by 61% overall. Being cited is a partial hedge against a decline you cannot otherwise avoid.
Published February 2026. Found by AI monitors how leading AI assistants answer buying questions, produces the content that addresses the gaps, and reports on what the published work earns - for agency partners, under their brand. See how the partner programme works.
Sources
- Similarweb, “Zero-Click Searches Surge to 69% Since Google AI Overviews Launched,” July 2025
- Previsible AI Data Study, analysis of 19 GA4 properties, May 2025
- Gartner, “Gartner Predicts Search Engine Volume Will Drop 25% by 2026,” February 2024
- Cloudflare, “The crawl-to-click gap: Cloudflare data on AI bots, training, and referrals,” 2025
- Superprompt, “AI Search Traffic Converts 5x Better Than Google: 2025 Conversion Data from 12M Visits,” 2025
- Search Engine Land, “Google AI Overviews drive 61% drop in organic CTR, 68% in paid,” 2025
- Semrush, AI Traffic Research, June 2025
- McKinsey, “New front door to the internet: Winning in the age of AI search,” 2025
- MDPI Journal of Electronic Commerce, “Consumer Responses to Generative AI Chatbots Versus Search Engines for Product Evaluation,” 2025
- Sam Altman via TechCrunch, ChatGPT weekly users reach 800 million, October 2025
- SparkToro, “2024 Zero-Click Search Study,” 2024
What to do with this
The gaps are the brief. We write what fills them.
Monitoring shows which buying questions the leading AI assistants answer with somebody else’s name. Those questions become the month’s articles, answer blocks, fact sections, FAQs and refreshes - written, published and measured under our partners’ brands.