Content teams need an evidence layer

Planning for an industry event this week forced me to name the job I was actually trying to do.

I wasn’t mainly trying to produce event content. There would be sessions, interviews, videos and the usual opportunities to turn a busy day into a collection of assets, but that wasn’t the interesting part. The useful opportunity was to work out what the organisation still didn’t know well enough, find people who could answer those questions from experience, and bring that evidence back in a form that could strengthen more than one piece of work.

My work AI gave that a useful description: evidence acquisition and evidence activation.

I think that is becoming a much bigger part of the content job.

AI has made production easier. It can turn a transcript into an article, an interview into FAQs, a report into social posts and a rough brief into reasonably competent copy in minutes. That does not make production unimportant, but it changes where the difficult part sits.

If organisations can all produce more words, the advantage moves further upstream. The scarcer question becomes: what do you know that is specific, credible and worth producing anything from?

Production is no longer the whole constraint

For a long time, content operations were built around production scarcity. Writing took time, editing took time, repurposing took time, and adapting one idea for several channels could consume most of a week. It made sense to organise teams around calendars, briefs, templates, production queues and the number of assets they could ship.

AI compresses a lot of that effort. A competent model can now produce the visible layer of content without necessarily doing the difficult work that once sat underneath it. That is the point I made in The words were never the whole value: the article was visible, but the research, judgement, interviews, exclusions and connections that made it worth reading were often hidden inside the word “writing”.

The important consequence is not simply that writers can work faster. It is that competent prose is no longer strong evidence that an organisation has done much thinking at all.

A page can be polished and still generic. A thought-leadership article can be fluent while containing nothing that another company could not publish. A customer answer can be perfectly structured while avoiding the operational detail that would make it genuinely useful.

AI has made the surface easier to produce. That makes the quality of the material underneath it more important.

The real bottleneck is often evidence

Imagine two organisations writing about the same business problem. Both can access the same public research, both can search the same internet and both have models capable of producing good first drafts.

One asks AI to write an article from those sources. The other has spoken to customers, interviewed practitioners, captured implementation examples, found where the standard explanation breaks down and kept the details that did not fit neatly into the original brief.

The difference is not writing capability. It is evidence.

A practitioner explaining what actually changed inside a workflow is evidence. A customer describing the point where a process failed is evidence. A support team seeing the same misunderstanding every week is evidence. An expert showing why the technically correct explanation is operationally useless is evidence.

AI can help retrieve, compare and synthesise that material. It cannot create credible first-hand evidence that nobody bothered to collect.

This is why I increasingly think content teams need an evidence layer underneath the production layer. I already describe the infrastructure version of this as the Evidence Engine, but the practical point is simpler than the framework: before asking what we should make, ask what we need to know and prove.

GEO makes the gap easier to see

GEO has made this especially obvious to me because visibility work quickly exposes the difference between a page problem and a knowledge problem.

Sometimes the organisation has a good answer and the page expresses it badly. That is an optimisation problem. The answer may need clearer structure, better retrieval, stronger entity signals or a more direct response to the question people are actually asking.

But sometimes the organisation does not have a strong answer yet.

You can see the question. You can see that customers care about it. You can see competitors being associated with it. You may even have a page that mentions it. What you do not have is enough evidence to say anything distinctive or useful.

At that point, rewriting the page is the wrong intervention. The missing asset is not another paragraph. It is an answer worth putting in the paragraph.

That produces a different loop:

question gap → evidence gap → evidence capture → usable answer → activation → new signal

This is why I think GEO should start with what your organisation knows, but it should not stop there. A useful visibility programme can also show you where the organisation needs to go and learn something new.

Evidence acquisition is an editorial capability

Once you see the problem this way, some old-fashioned journalism skills start looking unusually current.

Interviewing matters. Observation matters. Follow-up questions matter. So does the ability to recognise when somebody has given you the approved answer instead of the useful one.

Someone says AI saves their team time. The obvious move is to capture the quote. The evidence move is to ask what task disappeared, what replaced it, whether the work moved into checking and exception handling, and what happened to the recovered time.

Someone says clients are confused. Which clients? Confused about what? What happened the last time that confusion appeared? Who had to fix it, how long did it take, and was any of that work charged?

Someone says automation transformed the practice. What did a normal Tuesday look like before and after?

The first answer often gives you the message. The second or third answer gives you the evidence.

That is not a prompt-engineering problem. It is editorial judgement applied before the content exists.

Events are temporary evidence environments

This changes how I think about events too.

The familiar model is event → coverage → social posts → recap. There is nothing inherently wrong with that. If the job is event visibility, those outputs may be exactly what is needed.

But an industry event also puts practitioners, customers, experts, partners and competitors in the same place for a short period of time. That is a temporary evidence environment, and it creates a different question: what do we need to learn while these people are available?

I have written before about gathering the evidence you are missing rather than simply gathering content from the room. The more I apply that idea, the more important the distinction becomes. A useful event plan can start with known evidence gaps, use repeated questions to test whether a pattern is real, and capture examples that can strengthen work long after the event itself has disappeared from the calendar.

One conversation might improve an FAQ. Another might expose a weak assumption in a campaign. Three similar answers might justify a new article. A disagreement might tell you that the neat narrative you arrived with is too simple.

That is much more valuable than treating every quote as an output.

Then comes evidence activation

Acquiring evidence is only half the problem. Most organisations already have far more useful material than their published content suggests.

It sits in customer interviews, research reports, sales conversations, webinars, support tickets, event transcripts, product discussions, advisory boards, meeting notes and forgotten decks. The problem is that each piece of evidence tends to be consumed by the output that commissioned it.

The customer interview becomes the customer story, then disappears. The research becomes the launch campaign, then goes into the archive. The event conversation becomes the recap, then the next event begins with another blank document.

That is the problem behind Why content work needs to compound. Good editorial work should leave behind more than the published asset. It should leave behind knowledge the next piece can inherit.

Evidence activation is the work of making that possible. It means preserving enough provenance and context that the material can be found later, understanding which questions it answers, connecting it to the narratives or customer problems it supports, and knowing when it is strong enough to reuse.

The flow I am increasingly interested in is:

question → evidence → structured memory → interpretation → multiple future uses

That is deeper than repurposing. Repurposing turns one output into several formats. Evidence activation lets the same underlying knowledge produce different outputs because different future questions can interrogate it in different ways.

A useful evidence record can stay small

I do not think this requires building a giant knowledge platform before anybody is allowed to ask a customer a question. That would be exactly the kind of system-building that makes the work serve the infrastructure instead of the other way round.

A useful evidence record can be small enough to fit in a table:

Field	What it preserves
Question	What were we trying to understand?
Evidence	What did the person actually tell us?
Source	Who said it, and in what context?
Example	What concrete situation supports the claim?
Theme	What larger problem or narrative might this inform?
Confidence	Is this one anecdote, or part of a recurring pattern?
Reuse	Which existing answer, page, brief or campaign could it strengthen?
Gap	What do we still not know?

The important thing is not the schema. It is that the evidence survives its first use.

AI makes that much more practical than it used to be. Models can extract promising claims from transcripts, compare interviews, cluster repeated questions and help retrieve old material when a new brief appears. They can reduce the cost of maintaining context across a large body of work.

But the sequence matters. AI should help organise and interrogate evidence, not become a substitute for acquiring it.

Otherwise we end up with increasingly sophisticated systems for recycling information everyone already has.

There is a strong counterargument: content teams do not need another layer

I can see the objection to all of this. Many organisations already have research teams, customer-insight functions, CRMs, knowledge bases, content repositories and analytics platforms. Adding an “evidence layer” can sound like one more conceptual box sitting between people and the work they are trying to get done.

That objection is right if the answer is another platform, another taxonomy or another process people have to maintain for its own sake.

The real gap is often ownership rather than tooling. Someone has to notice that a recurring customer question is not properly answered. Someone has to turn that gap into a better interview question. Someone has to recognise that three separate conversations are pointing at the same operational problem. Someone has to decide that the evidence is strong enough to change the brief, or weak enough that the claim should stay a hypothesis.

Those decisions already happen. They are just often scattered across different people and lost between projects.

So I do not think the first move is to build an evidence platform. The first move is to take one live business question, identify what the current answer cannot demonstrate, gather a few strong records, use them and see whether the work improves.

If it does, preserve the pattern. If it does not, stop building the system.

The content job is moving closer to the source

This is the broader shift I keep seeing in my own work and in the roles appearing around the market. I have called it editorial work moving upstream, because the production layer is becoming easier while more value sits in the choices that determine what should be produced.

Evidence acquisition and evidence activation make that shift more concrete.

The future content professional is not simply the person who receives a brief and turns it into an asset. They may be the person who identifies what the organisation cannot yet prove, gets closer to the customer or practitioner who can explain it, preserves what was learned, and then helps that evidence travel through articles, video, campaigns, sales material, GEO answers and future briefs.

Writing still matters. Editing still matters. Distribution still matters. But they work better when they are downstream of something specific and defensible.

That is why I am increasingly less interested in asking how much content a team can produce. The more useful question is what the team is learning that makes the next piece better than the last one.

AI can produce more content from what an organisation already knows.

The harder job is making sure the organisation keeps learning something worth producing content from.