What usually decides the balance?

Whether the person would find the processing unexpected given how their data was collected. Surprise is a more reliable practical test than any internal judgement about how important the business need feels, because the second is assessed by the party that benefits from it.

Why do AI uses strain this ground?

Because analysing, classifying or generating from existing material is frequently something the person never contemplated when they handed it over. The business need is usually genuine, so an assessment tends to fail on the expectation half rather than on whether the organisation has a real reason for wanting it.

How thorough does the record need to be?

Thorough enough to show the weighing happened and what it concluded. A short document naming the interest, the effect considered and why one outweighed the other is worth considerably more than a long one that restates the activity without ever reaching a judgement.

Glossary

Legitimate interest

A ground relying on a genuine business need weighed against the individual's rights, where the weighing has to be done and written down.

In plain terms

Saying you have a good reason and that it does not unreasonably affect the person. What makes it different from the other grounds is that you have to demonstrate you thought about it, which means the reasoning is the thing rather than the conclusion.

Why it matters

Because it is the flexible ground and flexibility is exactly what makes it easy to misuse. It suits activities an organisation genuinely needs to keep doing, and it is also the ground somebody reaches for when no other one fits, which are two very different situations that produce the same sentence in a document.

How it works

It requires weighing your need against the effect on the person, and both halves have to be real. An interest that is genuine but trivial does not outweigh much, and a substantial interest can still lose against an effect somebody would find intrusive or unexpected.

The assessment is the substance and the record of it is the evidence. Unlike the other grounds, this one asks you to show your working, so an organisation relying on it without a written assessment has not really relied on it; it has asserted it, which is a different thing when somebody asks.

Expectation is what the balance usually turns on. Processing that a person would find unremarkable given how their data was collected sits comfortably; processing they would be surprised by does not, and surprise is a better test in practice than any judgement about how important the business need feels internally.

It cannot be used to reach something the person has actively objected to in most circumstances, which distinguishes it sharply from a ground the organisation simply declares. The individual retains a say, and an assessment that ignores that has skipped the half that makes the ground legitimate.

AI activities strain it in a specific way. Feeding existing material into a tool that analyses, classifies or generates from it is frequently something the person never contemplated, so the expectation test is where such a use tends to fail rather than at the question of whether the business need is real.

The two halves, and which one gets done

The structural reason the right-hand column goes missing is worth naming, because it is not laziness. Everyone drafting the assessment works for the organisation with the interest, so the left-hand column is the one they can complete from knowledge and the one they are motivated to complete well. The right-hand column requires imagining the position of somebody not present, and no part of the process supplies that person or their view. What that produces is a document that reads as thorough, argues its case competently, and never actually performs the exercise the ground is named for. The remedy is small and slightly awkward: have somebody state the strongest version of the objection before the conclusion is written, rather than after. For AI adoptions this matters more than usual, because the business need is almost always genuine and easily argued, so an assessment built only from the left will always conclude in favour, and the question that would actually have changed the answer was whether the person whose material it is would have expected any of this.

Seen in the wild

Assessing whether customers would expect their past correspondence to be summarised by an assistant.
ChatGPT
A search deployment surfacing material about people in contexts they would not anticipate.
Glean
An automation scoring or sorting individuals in a way nobody described when their data was collected.
Make

Common misconceptions

People assume

It means we can proceed if we have a good reason.

In fact

It means the reason is weighed against the effect on the person, and the weighing has to happen and be recorded. A good reason on its own is half the test, and it is the half that does not require anybody to think about the individual.

People assume

The assessment is paperwork.

In fact

It is the only evidence that the ground was relied on rather than asserted. Without it there is nothing to distinguish a considered decision from a preference, which is precisely the distinction anybody reviewing it will be looking for.

Questions

What usually decides the balance?: Whether the person would find the processing unexpected given how their data was collected. Surprise is a more reliable practical test than any internal judgement about how important the business need feels, because the second is assessed by the party that benefits from it.
Why do AI uses strain this ground?: Because analysing, classifying or generating from existing material is frequently something the person never contemplated when they handed it over. The business need is usually genuine, so an assessment tends to fail on the expectation half rather than on whether the organisation has a real reason for wanting it.
How thorough does the record need to be?: Thorough enough to show the weighing happened and what it concluded. A short document naming the interest, the effect considered and why one outweighed the other is worth considerably more than a long one that restates the activity without ever reaching a judgement.