AI in practice: Editorial strategy

Q: How are publishers using AI in editorial strategy?

A: Much of AI’s immediate value comes from helping editors organise content and reducing repetitive work. The first question publishers should ask is whether they actually want to improve: more consistent tagging, a more useful archive, or less time spent preparing articles for publication.

I’m not sure many publishers have made AI a routine part of their editorial strategy yet. There is still fear around getting things wrong in news, and many tools haven’t behaved reliably enough for editors to trust them.

There’s also a question of where the saved time should go. Faster production does not necessarily mean less work or mental load. An editor who once typed five tags might now need to review thirty suggested ones. The pace rises and so does the number of decisions. The editor still has to decide what matters, what is correct, and what should be published, and that judgement requires time.

So far, AI has mostly been added as a button, a chat box, or a separate assistant. I’m more interested in AI that works with the way a publication organises its content and makes editorial decisions, one that understands the publication’s tags, house rules, and editing process, so editors are not re-explaining them every time.

Some editors will want that tooling inside their publishing system; others will prefer their own assistant. Either way, publishers need to be clear about which editorial decisions they are willing to delegate and who remains responsible for the outcome.

Models and tools will keep changing, so the durable investment is in the groundwork: organising the archive, agreeing how content should be classified, and helping staff learn where AI is useful. That’s the work that survives and serves as your foundation, regardless of model or tool.

Q: What is a particularly illustrative use-case?

A: Taxonomy, the tags and categories used to organise content, is a good example.

An article might mention David Beckham many times while actually being about his children. A tagging system that ranks subjects by counting mentions could decide he is the main subject. But being mentioned most often does not necessarily equate to being the main focus of what the article is about.

Editors have to decide which distinctions matter to their publication. Those decisions are what a taxonomy is built on and give them a basis for judging AI suggestions.

A stronger approach is to combine several signals: exact matches, what the article is about, tags on similar articles in the archive, and tags that tend to appear together. Publishers can give those signals different weights to suit their taxonomy. Because each signal can be wrong on its own, agreement between them gives an editor more reason to consider a suggestion, though it still needs checking. If the suggestions are relevant seven times out of ten, that is already useful: the editor has a starting point rather than tagging every article from scratch.

Suggestions can also fill gaps. An editor may see a relevant tag they would not have thought to add, helping them classify the story more completely. Those accepted tags then become part of the archive used for future recommendations. Better tagging can therefore improve later suggestions, provided editors keep the data accurate.

For a publisher deciding where to invest, maintaining that archive and its taxonomy is part of the AI work, not overhead separate from it.

Q: What are your three top tips?

1. Start with an editorial priority. Decide what you want to improve before choosing a tool. That might be more consistent tagging or reducing repetitive work before publication. Give editors a specific task to try and judge whether the suggestions genuinely help. Where they keep supplying the same content or repeating the same steps may be something worth automating. Use that experience to decide where to invest next.
2. Develop some mechanical sympathy for AI. Editors and decision makers need direct hands-on experience with these tools. Use them enough to learn what they do well, where they become unreliable, and how they respond to instructions. That’s insight you can’t get entirely from demos or strategy documents. Once you understand what the tools can do and how your own publishing process works, useful applications become easier to spot.
3. Decide what AI can change or publish. Editors need to know what the AI can suggest, what it can change, and what still requires human approval. Those decisions should reflect the consequences of getting something wrong and be used to set the limits. There is room for more automation in drafts and test environments, so not every action needs to require a human click. But the publishing system, not the AI model or a cleverly phrased prompt, must enforce the agreed limits.

About us

Glide Publishing Platform is a SaaS suite purpose-built for media, publishing, sports, and entertainment companies. Glide’s product suite includes the agent-ready headless content management system, Glide CMS, and the audience interaction platform, Glide Nexa. Our customers include The Times, Daily Mail, Which?, HELLO! Magazine, Editorial Perfil, Arsenal, and AS Roma.

Email: sales@gpp.io

Web: www.gpp.io