Editor’s Pick Platforms & Discovery Editors Picks

The book trade’s discovery problem starts with its own data

Ivan Massow 8 Nov 2026 · 7 min read

A publisher in red files a book-information card into a large gold cabinet. A bookseller in a blue apron holds a book on the other side, reaching towards the cabinet’s inaccessible closed back. The caption reads: It’s all in the system.

Cartoon · AI-generated

NielsenIQ BookData’s new metadata study gives publishers an awkward sales comparison. In the 2025 UK print data it examined, titles with a long description sold 119% more copies on average than those without, according to The Bookseller’s account on 5 November. That deserves attention. It is not a promise that adding a paragraph will double a book’s sales.

The product record has become part of the shop window. Descriptions, subject categories, edition details and cover images help retailers show a book to the right customer. Readers gain when that information is useful; publishers and their partners carry the cost of keeping it correct. The decision is where incomplete or badly transmitted data is obstructing a sale, before spending on another discovery product.

Metadata is the structured information describing a book. The study joins BookData’s records to BookScan sales for the 100,000 bestselling printed titles of 2025. It compares groups with and without particular information, including keywords, which were associated with 113% higher average sales. This is a selected print-sales population, not a finding about every published book or the effect of changing one field.

“Readers cannot buy books they cannot find,” report author Philip Stone, head of publisher account management at NielsenIQ BookData, said, as quoted by Publishing Perspectives on 6 November. The argument is straightforward. The size of the difference needs more care.

Advertisement · Noah

Popular authors, larger marketing budgets, genre, age and availability could influence both the attention given to a record and the number of copies sold. Those are possible explanations to test, not findings established here. The publicly accessible accounts do not show how much of the sales gap a metadata improvement would cause. BookData also sells metadata services. Its commercial interest makes the comparison’s method and limits worth reading closely.

Good information also requires judgement. “Not every publisher uses Thema yet, and not all use it well,” EDItEUR standards expert Chris Saynor wrote in a July guest blog for the European and International Booksellers Federation. He also said adoption and coding quality were improving. Thema is a shared subject-classification system: a way to tell booksellers what a book is about without relying on one supplier’s vocabulary.

A separate Ingram case study offers a more useful intervention comparison. Published in April, it describes 2,000 print titles, priced in US dollars, published between 2019 and 2021 and previously selling 24 to 120 units a year. Ingram divided them into comparable genre-balanced groups, enriched one group’s classifications and keywords, then tracked sales after distribution. It reports a 9% advantage in US sales over a year and 22% in UK sales for the improved titles.

That is a different population and design from NielsenIQ’s snapshot. The account does not describe random assignment, and the work changed several metadata elements together. It cannot identify the return from one keyword or one supplier’s service. It nevertheless points towards a more informative question: what happens when information is deliberately improved, compared with similar titles left unchanged? Ingram is a distributor selling services, and its footnote says similar results are not guaranteed.

The record has to arrive

Consider a hypothetical guide to coastal birdwatching. A broad nature classification tells a retailer less than an accurate description of its subject, place and intended reader. More detail can help the shop organise or retrieve it, if the shop receives and uses that detail. Adding unrelated popular search terms would make a fuller record, but a worse description of the book.

The route from publisher to reader can run through distributors, data aggregators, wholesalers and retailers. BIC, the UK book industry’s supply-chain organisation, maps those relationships and recommends continuing updates before and after publication. Sending a correct file is one part of the job. Establishing what a trading partner accepted, changed or displayed is another.

Waterstones’ published trading pack makes the handover tangible. It identifies Nielsen BookData as its sole source of book information and directs publishers wanting to add a synopsis to the enhanced listing service. Basic records are free; enrichment is a subscription product. BookData’s own site says annual charges depend on the number of titles, with a minimum charge. Making a book orderable and supplying richer selling information are distinct tasks, with distinct costs.

BIC’s recipient guidance distinguishes a file that follows the ONIX messaging standard from one that contains correct information. It gives the example of a book still marked forthcoming despite a publication date in the past. Both fields can exist, yet together mislead the buyer. A more elegantly written synopsis would leave that fault untouched.

The same guidance warns that recipients imposing their own variants of ONIX can increase suppliers’ cost of serving them. ONIX is the agreed format for exchanging book information, not a guarantee that every recipient interprets or displays everything identically. The operational question is who owns the correction when the publisher’s record and the customer’s screen disagree.

This is a commercial handover, not simply a database exercise. A publisher can inspect a sample of its priority titles at the point of sale and compare the visible record with the approved one. Missing descriptions, wrong editions and stale availability need different remedies. An enrichment contract should make clear which of those problems it can actually solve.

Noah, a research and forecasting service, puts only a 24% chance on NielsenIQ BookData, Gardners or Ingram launching, after 8 November 2026 and by 8 November 2027, a new production service for UK trade publishers that gives ISBN-level acknowledgement, derived from a UK bookseller’s status, of whether it has ingested a corrected record. A distributor’s sending receipt would not qualify. Retailer cooperation and integration costs are obstacles; existing relationships could enable a narrow launch. Even ingestion would not prove that the customer sees the correction.

Earlier information, more responsibility

BIC’s current accreditation rules add urgency to that work. They specify complete metadata at least 16 weeks before publication for ebooks and three weeks for digital audio, with measurement scheduled from October 2026. The distinction recognises that different formats have different production timetables. A print-sales association should not be casually transferred into an identical expected digital-sales return.

The work is already established across publishers of different sizes. BIC’s July accreditation list includes Canongate, Nosy Crow, Pan Macmillan and Penguin Random House UK. That records participation in an assessed industry standard. It does not establish the sales return those businesses earned from meeting it.

Accessibility adds a different obligation: tell a buyer what the product supports. BIC has introduced supplementary Accessibility Tick requirements from October, covering specified digital-product information. Its documentation explicitly says the tick does not assess legal compliance. Missing the extra elements does not necessarily cause an application or renewal to fail overall; BIC intends fuller incorporation later, on a date still to be determined.

That distinction matters to a reader choosing whether an edition will work for them. A marketing claim about accessibility needs to match the actual file and its features. The record cannot make an inaccessible product accessible. Nor should an organisation-wide badge be treated as a detailed answer about every edition.

AI can help prepare descriptions, suggest terms and identify inconsistent records. The valuable test is how much verified work it removes. A fluent invented author detail or an inaccurate accessibility claim creates correction work downstream. Approved source material, product checks and a named person responsible for exceptions remain part of the cost.

For AI drafting, Noah puts the chance at 53% that at least one UK trade imprint owned by Penguin Random House, Hachette, HarperCollins, Pan Macmillan or Simon & Schuster newly makes generative-AI drafting of descriptions or subject keywords routine for its new UK print titles, with human approval before distribution, after 8 November 2026 and by 8 May 2027. For the same outcome by 8 November 2027, the estimate is 59.4%. Bounded copy tasks favour adoption; integration and checking costs can obstruct it. Pilots and pre-window adoption do not qualify. Neither estimate predicts a sales gain.

Test the sale, not the field count

The research gives a publisher a reason to examine neglected records. Repairs to wrong editions or availability should be distinguished from optional enhancements. A trial of the latter on comparable backlist titles needs to track price, promotion, stock and unusual publicity alongside the changes. Otherwise the attribution problem returns inside the company.

Verification should continue beyond successful delivery of a feed. Did the retailer show the changed description? Could a relevant reader find the correct edition? Did orders or conversion improve against an appropriate comparison, and did the gain cover preparation, checking and partner-support costs? A larger record and a larger sales figure are both incomplete answers on their own.

The awkward part is organisational. Editorial staff know the book; production knows the edition; sales knows the trading partners; a supplier may manage the feed. A fault can survive because every team has completed its own task. Responsibility needs to extend to the version the reader actually encounters.

The case for accurate, timely book information is strong enough to fund serious work. The case for booking a 119% uplift into a commercial plan is not established by the new comparison. Ingram’s intervention study supplies a reason to test, with its own limits. Publishers should follow the record all the way to the shop, correct what fails there, and measure what changes.

Research note: the new report’s figures are drawn from its public announcement and the cited trade accounts. The complete report was not inspected. Study limitations and the exact forecast definitions and ranges are in the linked research notes.

Sources

Source: Tomorrow’s Publisher