Why we stopped publishing every day

What the archive shows

We counted. The site holds 89 posts with a publication date, the first from 6 November 2025 and the most recent from 13 August 2026. Twenty-three calendar days carry more than one post. May 2026 alone carries 40, and 23 May carries nine.

Thirty-one of the May posts start from something published elsewhere: a Gallup, Deloitte or DDI figure, an HBR article, a tribunal statistic, a change in employment law. The template was the same each time: a hook published by someone else, a paragraph on what the research says, and the Sidestream service that addresses it. The other nine, all dated 23 May, are three topics written three ways each: a behaviour change workshop, a leadership programme and a team workshop, each in three variants aimed at three ways of phrasing the same search.

None of it was written by someone who had just run a session and noticed something. It was written to be found.

Google's own description of the problem

Google's spam policy defines scaled content abuse as many pages generated for the primary purpose of manipulating search rankings rather than helping users, and adds that this applies "no matter how it's created". Whether a script or a person writes the fortieth summary of the month makes no difference to the definition.

Its guidance on helpful content is a list of questions, and one of them reads like a description of our May: "Are you mainly summarizing what others have to say without adding much value?" Its newer guide on content for AI search separates commodity content, which "could originate from anyone", from content that carries a unique expert or experienced take, and names writing separate pages for every variation of a search as a route into the spam policy. That is a fair description of 23 May.

We are not claiming Google penalised us; we have no evidence either way. But when a search engine writes down what low-value content looks like and the description matches your archive, the archive is the problem.

The rule since 17 September

On 17 September 2026 we replaced the daily target with a gate. At most one post per calendar day, and only if it carries at least one element that only Sidestream can supply. If none exists on a given day, nothing is published. The internal wording is blunt: no post is better than an interchangeable one. Every number still needs a primary source we have opened ourselves, cited in the post. The gate names four elements:

• a design decision from a real scenario and why it was made, without client names or plot details that would let a reader recognise a client or reuse the scenario;
• an observation from delivery that a reader could not get from the source report;
• how something is measured in practice, including what it cannot show;
• a limitation or a failure and what changed because of it.

What the check turned up

Checks of the existing site in the same week found two things we would rather not have found.

The first was a figure. Pages of our own stated a per-employee training spend, attributed to a professional body's report. The edition cited had never been published, the edition that exists gives budget ranges rather than a per-employee figure, and a web search returned the number only on our own pages. It had been listed internally as verified. On 18 September we replaced it wherever it appeared with the official figure from the Department for Education's Employer Skills Survey 2024.

The second was a word. For nearly three weeks, from 29 August to 17 September, the site described co-founder Ben Laumann's 2020 UCL study with a word that implies a control group. It has none. It is a randomised study: 57 participants were allocated at random to one of three ways of learning a feedback model. There was no control condition, so it compares three teaching formats against each other, not against no training at all. The word was corrected across the site on 17 September, and a second pass the same day caught pages the first had missed because a link sat between the word and the name of the study. The research page now sets out what the study does not show alongside what it does.

What we publish instead

Here is the kind of finding the daily template would never have surfaced, because it makes our own evidence less tidy. In the same UCL study, participants rated their confidence on a scale of 1 to 7 at three points: before the intervention, after it, and after delivering a feedback conversation that two assessors then rated blind.

Confidence rose in every group, and the format made no significant difference to it (p = .958); the figures are on our research page. Blind-rated delivery did differ: Cohen's d was 0.75 for practice against slides and 1.08 against video. A satisfaction survey at the end of that study would have declared all three formats a success, including the two that produced measurably weaker delivery. That is an argument about how training should be evaluated, and not one a reader can get from a summary of somebody else's report.

The caveats hold and are part of the point. It is one study of 57 people, run in 2020. The practice condition was an online coaching session of at most 20 minutes, the two passive formats had the same time limit, and the study measured one skill once. It says nothing about the effect of our current programmes. It says a good deal about what a feedback form measures.

What this means if you read us

Frequency will be irregular, and the irregularity is the method working. A quiet fortnight means nobody ran a session, checked a figure or changed a design in a way worth writing down. It does not mean we have stopped. Pages we corrected were corrected in place, because removing them would hide the error rather than fix it.

If a figure on any page of this site cannot be traced to a source you can open, tell us. We will trace it or take it down. And if you want to see the kind of evidence that passes the gate, start with the research page; the new manager training page shows what we built on it.

Sources: Google Search Central, Spam policies: scaled content abuse · Google Search Central, Creating helpful, reliable, people-first content · Google Search Central, Optimizing your website for generative AI features on Google Search · Department for Education, Employer Skills Survey 2024 · Sidestream, The Study Behind the Method