BookSeeking
How it worksFor Publishers
About
Sign inExplore the demo →
For publishers

Reach the readers most likely to value your catalogue.

We're building a recommendation engine that understands what a book teaches, not just what category it sits in. Better book data means better matches between your titles and the readers who'd actually want them.

The opportunity

The industry produces noise. We produce signal.

Most systems match by genre, author or co-purchase — and suggest a fifth vampire novel to someone who's read four. We analyse what a book actually teaches, so when a reader is ready for one of your titles, we surface it with precision. We never retain your source text; we extract structured metadata and discard the rest.

What we'd ask of you
Full digitised text

For a modest set — roughly 500–2,000 high-value titles per year — processed once at ingest, for structured analysis only.

Bibliographic metadata

Title, authors, year, ISBN, BISAC codes, and original chapter structure where available.

A scope agreement

Structured analysis only: no redistribution, no display of source text to end users.

What you get in return
Better-targeted matches

When a reader's profile points to one of your titles, we surface it with precision — signal, where the rest of the industry increasingly produces noise.

Chapter-level signal

Full-text titles gain a chapter-by-chapter concept map on top of the book-level layer — the asymmetric reward for contributing text, and the only way to power "this exact chapter fills your gap" recommendations.

Granular demand data

See which concepts readers seek, which chapters draw the most interest, and which of your titles are over- or under-indexed against real reader appetite.

Discovery, not just transaction

We optimise for reader development, not screen time. Readers who keep growing tend to buy more books over time, not fewer.

A worked example

Genghis Khan and the Making of the Modern World

Jack Weatherford · contributed in full

From the full text, our pipeline produces nine generic concepts — steppe military doctrine, premodern relay networks, 13th-century geopolitics — each anchored to a UDC class, with the people, places, periods and events attached.

Plus a chapter-level map

Eleven chapters, each tagged with the concepts it teaches and how central they are. So when a reader has the broad strokes but not the courier networks, we can recommend the one chapter — “The Yam relay as administrative backbone” — instead of the whole book. That granularity exists only because the full text was contributed.

Data & privacy commitments

We treat publishers as partners, not data to be scraped.

We don't retain your text

Analysis runs in memory. Source text is fetched, processed into structured metadata, then discarded — never written to our database, cached, or backed up.

We store only structured outputs

Chapter lists, concept labels anchored to UDC, appeal factors, and short pipeline-written summaries. Never substantial extracts of your books.

Readers never see source text

The product surfaces concepts and rationales. Where a recommendation cites a chapter, it's by title and page number — not by excerpt.

Reversible by design

If a partnership ends, the structured metadata can be de-attributed and kept useful only for matching, or deleted on request.

Let's run a pilot.

Start with 50–250 titles. We process them, you review the analysis, and we share back which of your books are over- and under-indexed against real reader appetite.