How Books Get Discovered by AI: ChatGPT, Rufus & the Zero-Click Era

Last reviewed by Robert Prime, July 2026

Readers are increasingly finding books through AI assistants and AI-summarised search results rather than by clicking through a page of blue links. In 2026, a growing share of "what should I read next?" moments happen inside ChatGPT, Perplexity, Google's AI Overviews, or Amazon's shopping assistant Rufus — and those systems answer the question directly, often without sending anyone to a website. For authors, that shifts discovery from ranking on a results page to being cited in an answer.

This is a genuine change, but it's easy to over-hype. Let me give you the honest version: what's measurably happening, what it means for how books surface, and what is still speculation dressed up as strategy. I run publishing.co.uk and co-own the LoveReading.co.uk review network, so book discovery is what I watch for a living.

TL;DR — Key Takeaways

• "Zero-click" search is now the majority. By early 2026, around 68% of US Google searches ended without a click to the open web (SparkToro), up from the high-50s in 2024. AI Overviews reach over two billion people a month (Google).
• AI Overviews roughly halve organic clicks. Pew found that when an AI summary appears, users click a traditional result in about 8% of visits versus 15% without one, and click a link inside the summary just 1% of the time.
• AI referral traffic is growing fast but is still small. AI assistants sent a rapidly rising number of referrals through 2025–26, yet still account for only around 1% of total web traffic. Both halves of that sentence matter.
• Amazon Rufus is live in the UK (since September 2024) and answers shopping questions using listing details, reviews, and community Q&A. It was renamed "Alexa for Shopping" in the US in May 2026; UK still shows "Rufus".
• AI answers cite sources, but which sources is volatile. Across the general web, Reddit and Wikipedia dominate citations; recommendation queries lean on review sites and "best of" listicles. There is no rigorous study of what AI cites for books specifically.
• What changes for authors: discovery is shifting from ranking to being referenced. The durable response is old-fashioned — be genuinely present and well-reviewed across the places these systems read.

What "Google Zero" Actually Means

The term "Google Zero", coined by The Verge's Nilay Patel, describes the tipping point where Google answers questions itself instead of sending you elsewhere. We're a long way into it. SparkToro's clickstream research found that in the first four months of 2026, roughly 68% of US Google searches ended in zero clicks to the open web, up from the high-50s in 2024 (the two studies use different panels, so treat that as a direction, not a clean line). Google's AI Overviews, the summarised answer boxes, now reach more than two billion people a month.

The effect on clicks is measurable. A Pew Research study of real browsing behaviour found that when an AI summary appeared, people clicked a traditional search result in about 8% of visits, versus 15% when no summary appeared — roughly half. They clicked a link inside the AI summary only 1% of the time. Ahrefs, separately, found the top organic result's click-through rate was around 58% lower on searches that triggered an AI Overview (a correlation, on desktop).

For a self-published author, the practical translation is this: a reader searching "best cosy mysteries set in Cornwall" is now more likely to read a synthesised answer naming a few titles than to click through five blog posts and compare. If your book is named in that answer, you win. If it isn't, the reader may never reach a page where you could have ranked.

The AI Assistants Readers Now Ask

Search is only half of it. A rising share of book discovery happens inside conversational assistants, each of which now cites sources inline:

• ChatGPT reached around 900 million weekly users by early 2026 and added a shopping-research feature in late 2025 that reads and cites sources across the web.
• Perplexity built its whole product around inline, numbered citations, every claim links to where it came from.
• Google's Gemini / AI Mode surfaces sources alongside its answers and has been adding more inline links.
• Claude gained web search with inline citations in 2025.

The volume is worth keeping in proportion. AI referral traffic grew sharply through 2025 and into 2026, traffic to US retail sites from AI sources rose several hundred per cent year on year — but in absolute terms AI still drives only about 1% of total web traffic (Conductor, late 2025). It's a fast-growing slice of a very large pie. The reason to care isn't today's volume; it's the trajectory and the quality: several analyses find AI-referred visitors convert better than typical organic traffic, arriving pre-qualified after doing their research in the chat.

Discovery is shifting from ranking on a page to being named in an answer. The reader never sees the page you would have ranked on.

Amazon Rufus: The Assistant That Sits Next to the Buy Button

For authors, the most commercially relevant assistant isn't ChatGPT — it's Rufus, Amazon's generative shopping assistant, because it lives inside the shop. Rufus launched in the UK in September 2024 and is now available to all UK customers on the app and desktop. (Amazon renamed it "Alexa for Shopping" in the US in May 2026; UK shoppers still see "Rufus", so that's the name to use for a British audience for now.)

Amazon says Rufus answers questions using product listing details, customer reviews, and community Q&A, plus information from across the web. Amazon's leadership reported that around 250 million shoppers used Rufus in 2025, and that customers who engage with it are more likely to complete a purchase. A reader asking Rufus "I liked The Thursday Murder Club, what else would I enjoy?" is being handed a shortlist inside the buying flow.

Here's the honesty caveat, and it's important: Amazon has published no official guidance for authors or sellers on how to influence Rufus. Everything you'll read about "optimising for Rufus" (star-rating thresholds, keyword weighting, backend attributes) is third-party inference, not Amazon documentation. What Amazon does confirm is that Rufus reads your listing, your reviews, and your Q&A. That tells you where to put your effort without pretending anyone has cracked a formula.

What AI Actually Cites (and What We Don't Know)

If being named in an answer is the new discovery, the obvious question is: where do these systems get their names from? For the general web, several large studies (Profound, Semrush, Peec) agree that Reddit and Wikipedia are the two most-cited sources, with Perplexity leaning heavily on Reddit and recommendation-style queries pulling in review sites and "best of" listicles. But two cautions apply hard here:

1. Citation patterns are volatile. One study tracked ChatGPT's share of Reddit citations swinging from around 60% to 10% in six weeks. Any single snapshot is a moment, not a rule.
2. There is no rigorous study of what AI cites for books specifically. The reasonable inference (that book recommendations lean on Goodreads, Amazon, Wikipedia, Reddit, and roundup articles) is exactly that: inference from general-web data and industry observation. Claims that "OpenAI trained on Goodreads" are unverified. Be sceptical of anyone selling certainty here.

I'd rather tell you "we don't fully know yet" than sell you a tactic dressed as a mechanism. What we can say with confidence is which surfaces these systems demonstrably read, retailer pages, reviews, Wikipedia, Reddit, editorial coverage, and that's enough to act sensibly.

What Actually Changes for Authors

Strip away the hype and the shift is smaller and more familiar than the headlines suggest. The channels changed; the fundamentals didn't.

• Reviews matter more, not less. Rufus reads them. AI answers weigh consensus. A book with genuine, plentiful reviews is more "legible" to these systems than one with three.
• Being talked about across the web matters. If your book appears in genre roundups, on Goodreads lists, in Reddit threads, in reviews and interviews, you're present in the places AI reads. Ahrefs' large 2025 study found third-party mentions correlate far more strongly with AI visibility than backlinks do (correlation, not proof — but a strong hint).
• Clear, accurate metadata helps machines understand your book. Precise categories, an honest blurb, and comparison ("if you liked X…") language give both Amazon's search and AI systems something unambiguous to work with.
• Your owned channels get more valuable, not less. When discovery gets noisier and less clickable, a reader on your email list is a reader you can reach directly, no algorithm in between.

The trap to avoid is over-rotating. Don't rebuild your whole strategy around a channel that's 1% of traffic today. Do make sure that as it grows, your book is already present in the sources these systems read.

How to Improve Your Odds — What the Evidence Actually Supports

You can't make ChatGPT or Rufus recommend a specific book on demand — nobody has demonstrated a formula, and you should be sceptical of anyone selling one. But there is a clear evidence hierarchy for what helps, and it rewards substance over tricks:

• Strongest (experimental). The one causal, peer-reviewed result — the 2024 GEO study (Aggarwal et al., KDD 2024) — found that adding quotations (+~41%), statistics (+~34%), and citations (+~29%) to content raised its visibility in AI answers, while keyword stuffing hurt it. Treat the exact figures as directional; the direction is unambiguous.
• Solid but correlational. Ahrefs' 75,000-brand study (Dec 2025) found third-party mentions — brand and YouTube in particular — correlate with AI visibility far more than backlinks do.
• Hygiene, not a lever. Book/Person/FAQ schema helps machines parse your facts, but Google says it isn't required for AI answers and isn't a ranking factor. Add it; don't expect miracles.
• A documented dead end. Skip llms.txt — as of mid-2026 the major AI crawlers measurably ignore it (Google has said so; ~97% of published files get zero AI-bot requests).

A sensible checklist for a new author:

1. Get to a solid base of genuine reviews via a compliant ARC campaign (Rufus reads them). See our reviews economics guide.
2. Fix your metadata — categories, keywords, and a specific, honest blurb (how to write it).
3. Claim and tidy your Goodreads and Amazon author profiles with accurate data.
4. Build a simple author website with clear, quotable, factual pages and basic Book/Person schema.
5. Earn genuine third-party mentions — genre roundups, reviews, interviews, community participation.
6. Write comp-title language into your descriptions ("for readers of…").
7. Don't build an llms.txt, buy a "Rufus formula", or chase vendor citation multipliers.

It looks a lot like good, honest book marketing — because that's what it is. The AI layer rewards the same substance and presence that earned discovery before the assistants arrived. For our deeper reference on answer-engine optimisation, see the AI book discovery & AEO guide.

Common Mistakes and How to Avoid Them

Panicking about a 1% channel. AI discovery is growing, not dominant. Prepare for it; don't abandon what works now to chase it.

Buying "AI optimisation" packages. Nobody has a proven formula for getting a specific book recommended by ChatGPT or Rufus, because no such formula has been demonstrated. Treat sellers of certainty with suspicion.

Ignoring reviews and Goodreads. These are the surfaces AI most plausibly reads for books. Neglecting them was already a mistake; it's a bigger one now. See our guide to getting reviews.

Assuming SEO is dead. It isn't — it's changing shape. Being the source an answer draws from still requires the same substance that used to earn a ranking. Our measuring what works guide covers how to track referral traffic from AI sources as it grows.

Frequently Asked Questions

How do books get discovered by AI assistants like ChatGPT and Rufus?

AI assistants answer reader questions ("what should I read next?", "books like X") by synthesising information from sources they read, retailer pages, reviews, Goodreads, Wikipedia, Reddit, and editorial roundups, and naming specific titles, usually with citations. Amazon's Rufus specifically uses your listing details, customer reviews, and community Q&A. Getting discovered means being present and well-reviewed across those surfaces, though no platform publishes an exact formula.

What is zero-click search and why does it matter for authors?

Zero-click search is when a search ends without the user clicking through to a website, because an AI summary or answer box satisfied them on the results page. By early 2026, around 68% of US Google searches were zero-click. It matters because a reader who reads a summarised answer naming a few books may never reach the page where your book could have ranked — so being named in the answer becomes the goal.

Is Amazon Rufus available in the UK?

Yes. Rufus launched in the UK in September 2024 and is available to all UK customers in the Amazon Shopping app and on desktop. Amazon renamed it "Alexa for Shopping" in the US in May 2026, but UK shoppers still see the "Rufus" name as of mid-2026. It answers shopping questions using listing details, customer reviews, and community Q&A.

Should I change my whole marketing strategy for AI discovery?

No. AI still drives only about 1% of total web traffic, though it's growing fast and converts well. The sensible approach is to keep doing what works now (a converting listing, reviews, an email list) while making sure your book is present in the sources AI reads, so you're positioned as the channel grows. Don't rebuild everything around a 1% channel.

Can I pay to get my book recommended by AI?

Not reliably, and be wary of anyone claiming otherwise. There is no proven method to make ChatGPT, Perplexity, or Rufus recommend a specific book, and no advertising product that guarantees it. What demonstrably helps these systems is genuine substance, real reviews, accurate metadata, and third-party coverage, not a paid shortcut.

Does schema markup help my book get cited by AI?

Modestly, as hygiene. Adding Book and Person structured data helps machines read your facts correctly, but Google has stated schema is not required for AI answers and is not a ranking factor. Add it because it's cheap and tidy, not because it will win citations on its own. Vendor claims of large "citation multipliers" from schema are correlational and unproven.

Should I create an llms.txt file for my author website?

No. As of mid-2026 the major AI crawlers measurably ignore llms.txt — Google has said no AI system uses it, and large-scale analysis found almost no AI-bot requests for these files. Spend the time on reviews, metadata, and third-party presence instead.

About the Author

Robert Prime is a self-published author, veteran e-commerce strategist, and the founder of publishing.co.uk. Co-owner of the LoveReading.co.uk book-review network and founder of the Amazon growth agency MrPrime.com, he tracks how readers discover books across search, retail, and now AI. After navigating the marketing of his own book, Google. Panic. Repeat., he built publishing.co.uk to help UK authors keep up without the hype. He is a member of the Forbes Business Council.