What is Demographic targeting?

Demographic targeting restricts or weights delivery by attributes: age brackets, gender, and on some platforms parental status, household composition, income band or education.

The critical fact about all of these is how the platform knows them, and the answer is usually that it does not.

Some attributes are declared — supplied by the person when they created an account. Declared data is reasonably reliable where people answered honestly and where the account belongs to one person.

Most are inferred. A model estimates the attribute from behaviour, and like every model it has an error rate. Shared devices, household accounts, people who use a family member's profile and anyone whose behaviour does not match the pattern the model learned are all mis-assigned.

So a campaign restricted to an age bracket is not restricted to people in that bracket. It is restricted to people the platform believes are in it, and the gap between those two is invisible from inside the account.

Three cases, and they share a property: the restriction encodes something true about the business rather than something observed in a report.

Legal requirements. Age-restricted products and services, where advertising to a younger audience is prohibited. The restriction is not an optimisation and should be set conservatively.

Genuine unusability. A product a group cannot use — a service for a specific life stage, equipment with a physical requirement.

Licensing and policy constraints. Where a business is permitted to market to a defined group only.

Outside those, restriction is a hypothesis, and it is usually a costly one.

The pattern is familiar. A report shows one bracket converting better. The account excludes the others. Costs improve briefly and volume falls, and the improvement turns out to be arithmetic rather than a gain.

Three reasons this happens.

The cells are thin. Splitting an account's conversions across several age brackets and two or three gender categories leaves most cells with too few events to distinguish from noise. A difference that looks decisive on a chart is frequently a handful of conversions.

The data is inferred. The comparison is between groups the platform assigned, which means part of the difference is model error rather than behaviour.

The model was already using it. An automated bid strategy varies its bids on signals that include demographics, at a far finer grain than a bracket. Restricting on top removes people the strategy was already declining to overpay for, and also removes the ones it correctly valued.

The net effect is usually a smaller audience with more frequency against it, and a cost improvement that does not survive the next quarter.

Advertisers in regulated categories — housing, employment and credit being the most commonly named — face restrictions on how demographic and location targeting may be used, enforced by the platforms through a declared-category setting. Meta's is called Special Ad Categories; Google Ads applies its own restrictions to the same verticals under a different name.

These constraints are not a matter of judgement and they are not negotiable within the platform. An advertiser whose product falls into one of these categories should establish the constraints before building the campaign, because the alternative is a campaign that is refused, disapproved, or approved and later found to be non-compliant.

The categories and their rules differ by platform and by market, which is an argument for checking rather than assuming a rule learned elsewhere applies.

Let the creative do the selecting. An ad that speaks to a particular person selects its own audience twice: directly, because people who recognise themselves respond, and indirectly, because the platform learns from who responds and shifts delivery accordingly. This works better than a bracket and it does not depend on the platform having guessed correctly.

Use exclusions for facts, not for patterns. A restriction that encodes something true is durable; one that encodes a report is not.

Keep the inclusions broad. Delivery models find demand in places a manually chosen demographic would have excluded, and the cost of leaving them room is generally lower than the cost of the restriction.

Read the breakdown without acting on it. Demographic reporting is genuinely informative about who is responding, which is useful for creative decisions, for product decisions and for understanding the business. It is a poor basis for a targeting change and a good basis for a conversation.

Demographic targeting is one of the oldest controls in advertising and one whose usefulness has narrowed most.

It remains essential where the law or the product requires it. It remains informative as a report. And as a performance lever it has largely been superseded by delivery models that use the same signals more precisely, and by creative that describes its audience better than a bracket can.

Most platforms report an unknown or undetermined bucket alongside the named demographics, and it is frequently large.

It contains people whose attributes the platform could not determine: no declared data, insufficient behavioural signal, or a context where inference is unavailable. Its size varies by market and by platform and it is rarely negligible.

Two consequences follow. A campaign restricted to named brackets excludes the unknown bucket by default on some platforms, which can remove a substantial share of the reachable audience without anybody intending it. And a demographic breakdown that omits the unknown row is describing a subset while appearing to describe the whole.

Where a restriction is genuinely required, the treatment of unknown is a decision worth making explicitly: including it reaches people who may not qualify, excluding it forgoes people who probably do.

Read rather than acted on, demographic reporting earns its place.

It says who is responding, which is information about the product and the message rather than about the media. A business discovering that its customers skew differently than assumed has learned something worth taking to the people who make the product, not only to the people who buy the ads.

It surfaces creative mismatches. An ad performing well in one group and poorly in another is usually speaking to one of them, which is a prompt to make a second ad rather than to exclude the second group.

And it validates or contradicts the assumptions in a media plan, which is worth knowing even when no setting changes as a result.