Week 6 · the fourth dimension

Recommendation: the outcome you can't chase directly

Being recommended is the goal — and the one dimension you can't work on head-on. You reach it sideways, through the others.

Makefield · Berlin

The fourth question

From described to chosen

The first three dimensions get you into the answer: the model recognises you, describes you accurately, and frames you against the right peers. Recommendation is what they build toward — the moment the assistant doesn't just mention you but names you as the answer: "go with X." It's the fourth dimension, and it's the one that actually moves a buyer.

It's also different in kind from the other three, and the difference is the whole point of this lesson.

An outcome, not a lever

Why you can't optimise it head-on

Recognition, accuracy, and framing are things you can work on directly — you can make your account clearer, correct what's wrong, improve how you're positioned. Recommendation isn't like that. It's an outcome that emerges from the others plus the machinery underneath them. There's no setting that makes a model recommend you. Push on "get recommended" directly and you find nothing to push.

So you reach it sideways. You raise recognition, fix accuracy, strengthen framing and the sources behind them — and the recommendation follows, or it doesn't, but it follows from those. This is why the method spends most of its effort on the inputs: they're the part you can actually move. Treating recommendation as the lever instead of the outcome is the most common way this work goes wrong.

The honesty line

Anyone who guarantees it is selling

Because recommendation is emergent and the engines are not under anyone's control, no one can guarantee an AI recommendation — and the credible players say so plainly. It's worth noticing that even the most aggressive agencies in this space stop short of promising guaranteed citations or placements. If someone promises you a guaranteed recommendation in ChatGPT, that's a claim the mechanics can't support.

What honest work offers instead is eligibility: making you the kind of entity a model can recognise, trust, and reach for — and then measuring whether the answer actually moved.

It's plural and it drifts

There is no single recommendation to win

One more reason recommendation can't be chased like a ranking: there isn't one of it. As the earlier dimensions showed, different engines recommend different names, and even the same engine will answer differently on another day. "Are we recommended?" has no single answer — it's a distribution across engines and across time. Which means the only honest way to claim progress is to measure recommendation the same way, repeatedly, per engine — never to point at one lucky answer.

The halfway mark

The outcome half is done — now the machinery

That closes the first four dimensions — the outcome an AI answer produces: recognised, accurate, fairly framed, recommended. They're what a buyer experiences. But you don't change them by wishing; you change them through the machinery underneath — how clearly you're documented, how readable your information is to a machine, and whose sources the model trusts. Those are the next three dimensions, and they're where the hands-on work lives.

Next — Documentation: writing so machines, not just people, understand you. New here? Start with the overview.

Common questions

Questions this lesson answers

Can I make AI recommend my company?

Not head-on. Recommendation is an outcome, not a setting you can turn on — there's no control that makes a model name you as the answer. You reach it sideways: by raising recognition, correcting what's inaccurate, and strengthening how you're framed and the sources behind you. Do that work and the recommendation follows, or it doesn't, but it follows from those inputs. That's why the method spends most of its effort on the parts you can actually move. Treating "get recommended" as the lever instead of the outcome is the most common way this work goes wrong.

Can an agency guarantee my company gets recommended by ChatGPT?

No — and anyone who promises it is selling something the mechanics can't support. Recommendation emerges from the other dimensions plus the machinery underneath, and the engines aren't under anyone's control, so a guaranteed recommendation or citation isn't something that can be delivered. It's worth noticing that even the most aggressive agencies in this space stop short of promising guaranteed placements. What honest work offers instead is eligibility: making you the kind of entity a model can recognise, trust, and reach for — and then measuring whether the answer actually moved.

If I fix my recognition, accuracy, and framing, will I definitely get recommended?

It makes you eligible, not guaranteed. Improving the inputs raises the odds that a model reaches for you, but the recommendation is downstream of them and of the machinery you don't control, so it follows from the work rather than being promised by it. The honest way to know it happened is to measure — check whether the answer actually changed, per engine, rather than assume the fix produced the outcome.

Is there a single 'AI recommendation' I can win or track?

No — there isn't one of it to win. Different engines recommend different names, and the same engine will answer differently on another day, so "are we recommended?" has no single answer — it's a distribution across engines and across time. The only honest way to claim progress is to measure recommendation the same way, repeatedly, per engine, rather than point at one lucky answer.