The UK Publishing Industry in 2026
Last August, I published a piece of research on the size and scope of the UK book and journal publishing industry. It was quite popular on LinkedIn and continues to be one of the highest performing pages on this website. So, a year on, as one of my summer projects, I wanted to refresh the research using fresh data and tools. That refresh tells two parallel stories: an industry with more entrants competing for an essentially static pot, and the collapsing cost in time and money of the analysis sitting behind that conclusion. Both are critical lessons for publishers in 2026.
Methodology
Companies House makes data on UK-registered businesses available via API or .csv downloads: I wanted to work with the latter so I had a saved snapshot of the data to compare with 2025. For the first iteration of the research, I used ChatGPT o4-mini-high to develop Python scripts to work with large .csv files. I wanted to keep the same approach this year: even setting aside the size of the data, a script is a deterministic approach and therefore replicable by me or other researchers, whereas native analysis by an LLM would be less reliable.
This year I am using Claude rather than ChatGPT as my daily driver, and I was curious to see how much difference agentic AI tools would make. Last year, including some allowance for trial and error, the process took just under three hours of my time to compile the data (analysis and writing up the results took longer). Yesterday afternoon, I put a two hundred word prompt into Claude Cowork running Fable 5 High, giving it access to the blog post describing methodology and a local folder containing all of last year’s code and data. I gave a clear instruction that it should ask me questions to resolve any uncertainties, that it should include rigorous data verification steps, and that the 2025 execution was the floor, not the ceiling for the exercise. Claude asked what I thought were well-considered questions about how to handle the 2026 SIC code revision,1 whether it would be acceptable to use the ONS Postcode Directory for offline geocoding (a change of approach from last year but also more efficient), and what sorts of longitudinal analysis I would be interested in. Preparation took about ten minutes. I set it running and went to make a coffee.
Two minutes in, it reported that it couldn’t reach the Companies House download page directly and asked me to download the zipped .csv archive into the working directory, which I did—my final intervention in the process. Twenty-one minutes later, it had completed its analysis, updating and running the Python code in a sandbox. As part of the process, it had cross-checked the 2025 numbers using the new version of the script and confirmed no variance from my original calculations. It also caught an inconsistency in the data using grep and corrected for it. The output was a multi-tab Excel sheet containing a readme, the overall figures and thirteen tabs of specific analysis for clients. It also produced a Markdown file with descriptive statistics and draft, bullet-point commentary, though I should emphasise that everything below was written by me afterwards. Total time elapsed, thirty-three minutes, of which just seventeen needed my active involvement.
I’ll say more about the data below, but I was staggered at the speed and accuracy of the analysis. Clearly it helped that I was building on last year’s approach and retained data, but even if I had been starting from scratch, I don’t think it would have taken significantly longer: maybe an extra fifteen or twenty minutes of dialogue with the model to scope and plan. And, bluntly, I’m an MBA/generalist, not a developer or data scientist. Not having to use Python and a command line directly was a bonus, and validating that this worked in Claude Cowork means I can think about creating scheduled workflows for this task in future (think monthly deltas). The collapse in time taken meant that I could spend a greater proportion of the time I had allocated to the project on validation and analysis—and still save time overall. And that’s the whole ballgame. This pattern of research and analysis is the thesis of my book The Intelligence Multiplier, due to be published in December, but even I was surprised by the magnitude of the time saving in this case.
Population Analysis
I’m not going to explain every figure or baseline from last year: the same structural notes apply, and you can find that explanation and analysis here if you’re interested. Instead, I’ll concentrate on changes and trends. However, one overarching caveat is that it’s hard to get an exact handle on the size and scope of the industry because any small publisher that operates using the sole trader legal structure rather than a limited company does not appear on the Companies House data,2 and in the other direction there are publishing-adjacent businesses (independent authors, for example) or even non-publishing businesses that are registered with erroneous SIC codes. So it’s best to regard the following as directional.
The Companies House data identifies 12,435 publishing companies (+2.3% to 2025), of which 11,778 are active (+2.9%). The population of journals publishers is essentially flat, with the growth coming from books. The distribution of those publishers by size is unchanged in shape: a pyramid. Basically the entire net growth is in the thousands of micro-scale companies that make up the base of that pyramid, and the number of publishers filing anything larger than small-company accounts remains less than 150 in a population of more than 12,000.
| Region | 2025 snapshot | 2026 snapshot | Change (%) |
|---|---|---|---|
| London | 4,777 | 4,940 | 3% |
| South East | 1,593 | 1,596 | 0% |
| East of England | 984 | 1,033 | 5% |
| South West | 871 | 888 | 2% |
| West Midlands | 646 | 670 | 4% |
| North West | 632 | 650 | 3% |
| Scotland | 522 | 545 | 4% |
| East Midlands | 425 | 432 | 2% |
| Yorkshire and The Humber | 436 | 416 | −5% |
| Wales | 277 | 311 | 12% |
| North East | 184 | 196 | 7% |
| Northern Ireland | 91 | 89 | −2% |
One of the findings from last year’s research was that there was not a single MP without a publishing business registered in their constituency: the 2026 update suggests that holds for 648 of 650 MPs—an important message at a time when the government’s slogan is “good growth in every postcode”, cleverly interpreted by my colleagues at the Independent Publishers Guild for their autumn conference as “publishing growth in every postcode”. Publishing registrations are spread across the country, but there is still a clear metropolitan bias: overall 42% of publishers are based in London, up marginally on last year, and the picture is even starker among the assumed largest publishers (based on accounts filing type) where more than two thirds are headquartered in the capital—though as I noted last year, many also have satellite offices around the country. The strongest year-on-year growth was in Wales (12%) and the North East (7%), albeit off low bases.
2,319 companies registered as publishers were added in the last year, with 2,034 exiting, for a net change of +285. Most of those exits were entirely foreseeable from factors like a strike-off proposal in last year’s data, dormant status or never having filed accounts. Of the new entries, 1,900 were founded in the last year and 419 were existing companies that added a publishing-related SIC code this year. Looking at this from another angle, company formation is continuing to grow: calendar year registrations of publishing businesses were 1,730 for 2025 against 1,530 for 2024—up 13%. However, we need to be aware of significant attrition and survivorship bias: for example, the cohort of publishers founded in 2024 shrank from 1,530 on the register last year (1,424 active) to 1,031 this year, a 33% attrition rate. It’s probably never been easier or cheaper to start a publishing business, but it’s hard to sustain one.
So more publishers, but competing for a market that has not grown significantly in real terms. According to the Publishers Association, the industry grew from £7.2 billion in 2024 to £7.4 billion in 2025—growth of just under 3% over a period where UK inflation was over 3%, so the real terms impact was a small contraction rather than growth. In the long run, these figures are even starker, but it highlights a nuance that I didn’t explore properly in last year’s analysis. In adjusting nominal to real terms growth, one has to pick a deflation measure and publishing is a complicated industry for that. CPI is convenient, but tells a story about UK consumer prices. It doesn’t as accurately address institutional versus consumer sales, or exports—huge for UK publishing—which should really be deflated using the appropriate measure for the country of sale. There’s no single deflator that adequately addresses a basket of products and services including a paperback sold in the UK, a digital collection licensed to universities and a significant rights sale in USD. So, like the business population issue above, we’re left with a directional conclusion: up a little in nominal terms, down a little in real terms.
Next Steps
I started with the parallel stories of publishing growth and the changing shape of analysis. The common factor to both is that barriers to entry are dropping away. Anyone looking to replicate my work could do so with a Claude subscription and a lunch hour. As I’ve written previously, the value challenge with AI is not so much in doing the work, but in defining the job to be done. The next stage of this work, which I will write about in my next post on the subject, is taking commodity data and commonly available analytical tools and adding specifics: segmentation by size, professional membership, type of publishing and—one hopes most valuable of all—twenty-nine years of industry knowledge and judgement. If that richer dataset feels interesting but not something you could do in a lunch hour, please do get in touch.
-
This turned out to be academic, as the revised codes were not present in the data—but we didn’t know that going in, so it was helpful for Claude to identify the potential issue. As I’ve written previously, as and when the new codes are in use, it will be possible to break down commercial and academic publishing more clearly. ↩
-
I acknowledged this as a theoretical possibility in last year’s post, but thanks to Richard Charkin who wrote to comment that it applied to his own business, Mensch Publishing and confirmed the hypothesis. ↩