Human in the loop

AI does the work. A person takes the responsibility.

We are not anti-AI. We use it every day, and it is why a small studio can do this much. We are against AI nobody checked.

The rules

Six rules we do not bend.

01

Nothing ships unread.

If a person has not read it start to finish, it does not leave the building. That includes reports, pages, emails and alt text.

02

Every fact is checked.

Models are fluent, not accurate. Any number, name, claim or quote is verified against a source before it goes near your brand.

03

A name goes on it.

Each piece of work is signed off by a named person you can talk to. Accountability is the part AI cannot do.

04

Your voice, not the average.

We start from your customers’ language and your opinions, so the result sounds like you and not like everyone else’s prompt.

05

You know where AI was used.

No mystery. You can explain the process to your CEO, your board or your customers, because we explain it to you first.

06

The human can say no.

A reviewer who cannot reject the output is a rubber stamp. Ours can, and do, send work back to be done again.

The loop

Five steps. Two of them are a specialist.

  1. You

    Brief

    You tell us the goal, the audience and what good looks like.

  2. AI

    Heavy lifting

    Crawling, checking, researching and first drafts, done in minutes rather than days.

  3. Expert

    Check and edit

    A specialist verifies the facts, cuts the noise and rewrites in your voice.

  4. Expert

    Sign-off

    A named person puts their name to it. No sign-off, no delivery.

  5. You

    Approve

    You get work you can forward to your CEO without reading it twice.

Who does what

An honest division of labour.

The more a task depends on judgement, truth or accountability, the less of it the machine gets.

TaskThe AIThe human
Crawling a site and running checksDoes itRemoves false alarms, ranks what matters
Researching what your market searches forGathers and groups the dataDecides which searches are worth winning
Writing an articleOutline and first draftRewrite, fact-check, voice, sign-off
Design and visualsQuick routes to react toThe choices, the craft, the brand
Positioning and strategyA sounding boardAll of it
Anything legal, financial or a promiseNothingAll of it

What “human in the loop” actually means

What the check changes

Avoiding looking like your competitors.

The same tools with the same prompts produce the same pages. The edit is where a business starts sounding like itself again.

Straight out of the model

In today’s fast-paced world, businesses need innovative solutions. At Northfield Precision, we leverage cutting-edge technology to deliver best-in-class results that unlock your full potential. Our passionate team is committed to excellence in everything we do.

  • Could be any company
  • No facts to check, so none are right
  • Your competitor’s site says the same

After a DGTL editor

We machine aluminium and titanium parts for aerospace suppliers, in batches from one to five hundred. Most quotes go back the same day. If a drawing can’t be made as drawn, we say so before you order, not after.

  • Only this company could say it
  • Every claim checked with the client
  • Shorter, and it answers the buyer’s question

Illustrative example. Northfield Precision is not a real company.

Questions

Reasons for having humans in the loop.

Isn’t a human in the loop just a slower, more expensive way of using AI?

It is slower than pressing publish on a first draft, and much faster than doing everything by hand. The check is where the value is: it is the difference between forty pages nobody reads and four that bring in enquiries.

Why not let AI do everything and save the money?

Because AI is confidently wrong often enough to matter. It invents facts, misses context only your business knows, and writes to the average. The cost of one wrong claim on your website is higher than the cost of someone reading it first.

What does the AI actually do in your process?

The heavy, repetitive work: crawling sites, running checks, gathering search data, summarising research and producing first drafts. That is what lets a small studio do the work of a much larger one.

And what does the human do?

Everything that needs judgement: deciding what matters, checking facts, removing false alarms, writing in your voice, and putting their name to the result.

How do I know a person really checked it?

Each piece of work names the person who signed it off, and you can talk to them. If nobody is willing to put their name to something, it does not go out.

Will our content end up sounding like everyone else’s?

That is exactly what an unchecked model produces, because it writes the most likely next sentence. Our editors start from your customers’ language, your opinions and the specifics only you know, which is what a model cannot supply.

Is our data safe?

Ask us, and any supplier, three things: which tools are used, what data goes into them, and who can see it. We answer those in writing before any work starts.

Do you tell us where AI was used?

Yes. You should never have to guess. We would rather you could explain the process to your CEO than have it look like magic.

Next step

Time for some support?

You do not have to do all of this alone. Tell us what is on your list, and we will tell you honestly where we can help.