Editor’s Pick Audiences Editors Picks

Publishers should rebuild comments for the readers who never post

Ivan Massow 26 Oct 2026 · 6 min read

A red-shirted editor speaks through a huge trumpet while a reader in blue braces against the blast, contradicting the caption, ‘We’re ready to listen.’

Cartoon · AI-generated

The Financial Times tried a small intervention with a large ambition: ask readers a better question before they reach the comments. Its account reports more people opening the discussion, including readers who had ignored it the previous month. That is a useful starting point for publishers trying to build a reason to return.

This week’s publisher playbook from Discourse, the community software company, makes the opportunity more immediate. Automated spam checks and easier integration promise less friction. Its push towards discussions that AI tools can read adds another audience to consider. The priority should still be the person paying attention to the journalism.

Publishers should fund a focused experiment in useful reader exchange. Readers gain somewhere to test an idea or contribute experience; publishers gain a chance to build a direct relationship. Editors and moderators carry the cost. The decision is whether that exchange is valuable enough to deserve their time—not how cheaply a website can collect replies.

The invitation matters more than the box

Advertisement · Noah

In the FT experiment, an AI system suggested three questions based on an article. The editorial community team chose or edited one, placed about two-thirds of the way through the text. Human review remained part of the process. The aim was to make an invitation easy to produce without handing editorial judgement to the software.

FT senior product manager Bella Cockarill reports a 3.5% rise in comment views. She also says 11.5% of readers who had not viewed comments in the previous month did so that month. Those are participation results, not subscription gains. The undated account supplies neither the sample and comparison details nor a numerical retention result.

The useful lesson is about product design: put a relevant reason to join the discussion where an interested reader will see it. A generic invitation to have an opinion asks the reader to do all the work. A specific question gives the exchange a purpose.

Design for the reader who rarely speaks

Noah puts a 64% probability on at least one UK or European subscription news publisher launching, by 30 April 2027, a workflow in which editors invite selected readers to respond, humans review their contributions, and those responses appear in an article discussion. That would take participation beyond an open comment box towards a deliberately hosted exchange.

The ingredients already exist: editorial community teams, article prompts and human moderation. The obstacle is the work of selecting people and reviewing replies. Curating only familiar contributors could also narrow the conversation. The forecast favours a qualifying launch, not a guaranteed commercial success or an industry-wide change.

An illustrative example makes the difference clear. A local publisher covering bus cuts could ask affected passengers what a proposed timetable would prevent them doing, then review and surface useful answers. A reader who never posts could still find the discussion valuable because it contains experiences missing from the original announcement.

That quiet reader belongs in the product plan. Measure whether people read useful contributions and return for another visit, as well as whether they write one. A busier thread can reflect a handful of prolific participants rather than a better service for everyone.

There is no single access model. The Independent requires a free account to post and describes a 24-hour moderation team. Future’s woman&home announced registered-reader commenting in July 2025. These examples establish publishing choices, not proof that either approach improves subscription economics.

AI can read the conversation. Will people come back?

Noah assigns only a 34% chance that, by 31 January 2027, at least three separately owned UK or European news publishers will offer public Markdown versions of both discussion-topic lists and individual threads. For the same milestone by 26 October 2027, the estimate rises to 55%. That suggests a plausible, uneven deployment path rather than an imminent industry standard.

Markdown is a clean text format: software can read the conversation without working through the surrounding page design. Discourse explained its approach on 25 October, following a staff announcement on 22 September. Hosted sites receive the change through its update system; self-hosted installations need updating. Administrators can opt out.

The longer window allows routine upgrades to do some of the work. Against that, publishers may use different comment systems, keep discussions behind authentication or decline public machine-readable access. Three independently owned publishers is a small technical milestone; it is not a measure of market adoption.

For a publisher, the strategic question is what happens after the discussion becomes easier to consume elsewhere. A useful answer could help someone discover the title. It could also satisfy their question without a visit. Neither outcome is established merely by making the text accessible.

Insist on a return path

Noah assigns only a 41% chance to a UK or European news publisher publicly disclosing, by 26 October 2027, a numerical measure of visits, registrations, subscriptions or revenue attributed to identified AI assistants reaching its reader-discussion pages. This is a forecast about evidence being shared, not the odds that AI traffic exists or is worthwhile.

A public case study could meet that test with a small number of visits; it would not have to demonstrate profit. Conversely, a publisher could see useful traffic and keep its analytics private. The estimate therefore argues for measuring the channel carefully, not rejecting it.

Before widening access, decide what a successful return looks like: a reader following the source link, registering, exploring another story or coming back later. Check which of those actions can actually be attributed. The case for machine-readable discussion should be accountable to reader value, rather than resting on the appeal of being discoverable.

Prove the subscription case before scaling

Noah gives only a 26% probability that, by 26 October 2027, a news publisher serving UK or European readers will publicly release controlled or quasi-experimental research with a numerical subscriber-retention result for article discussion. That result could be positive, negative or negligible. The forecast concerns qualifying evidence appearing, not whether comments help people stay subscribed.

There is a useful warning in older research. The Center for Media Engagement randomly assigned 24 Gannett newsrooms to commenting arrangements in a study beginning in December 2019. Removing comments reduced time on site relative to retaining Facebook comments. Switching systems could improve discussion engagement, but the researchers detected no subscription-rate differences across the retained commenting systems.

That historical US experiment does not settle the question for a British publisher today. It does show why attention, participation and subscriptions need separate tests. Enthusiastic readers may comment because they already value a publication; their loyalty does not prove that asking everyone else to comment will reproduce it.

Run the first pilot on a beat where reader experience can improve the journalism. Give an editor responsibility for the question and a moderator responsibility for the queue. Compare a suitable invited group with a control group, count staff time, and follow the readers long enough to assess the outcome being claimed. Expand when useful participation and reader behaviour justify the cost.

The opportunity is editorial before it is technical. A well-hosted conversation can give readers something worth finding together. The investment earns a larger budget when the publisher can show what that experience adds.

Research note: Noah completed all 35 planned research checks across 23 distinct lines of enquiry. The archive results included off-topic matches, which were excluded from the commercial argument. Direct source checks and modelled forecasts inform this article; the full forecast register, uncertainty ranges and claim audit are linked below. An additional check of crawler costs, participation selection and historical chronology failed, so no further finding is claimed from it.

Sources

Source: Tomorrow’s Publisher