X and Twitter advertising targeting, the four filters X applies as AND and the ones that widen your audience instead

X advertising targeting splits into two kinds and getting them the wrong way round is the most expensive mistake in the account. Geo location, gender, language and age are what X's own help documentation calls subtractive, or AND, targeting. Every one of them is respected, and they stack. Interests, follower look-alikes, keywords, post engagers and uploaded audiences are additive, or OR, targeting, which widens the pool. X's own worked example, read on 31 October 2026, is blunt about the first kind. Choose the United States plus males plus 18 to 50 and "your campaign impressions will only serve to males 18 to 50 years old who are in the United States", whatever else you add. So demographics are the floor of your reach, and everything else is a request.

The four subtractive filters, and what each is built from

• Geo location. X documents targeting by countries, regions, metros, cities or postal codes, entered under the Locations section of the ad group. The caveat matters more than the list: "some countries do not have granular targeting options (metro, postal code, region) available at this time", and X's advice when a location does not populate is to add the closest region or area instead. So a campaign plan built around postal codes in one market may not be portable to the next.
• Gender. A straight filter, and like the rest it is always applied.
• Language. X publishes a long list of supported languages for targeting, which is the field most often left at its default when it should not be.
• Age. Set under Demographics as either all ages or a minimum and maximum. X states that age comes from the birthdates people entered on their profile, and that for users who have not provided a birthday it "extend[s] our age inference to these users based on a number of attributes such as the accounts they follow and their interests". That is worth knowing before you run a campaign with a narrow age band and treat the delivery as exact.

One geo behaviour catches people out and it is documented. If you target both a country and a region inside it, the ads show to the whole country. X's words are that your ads will show to all locations selected even when they come from the same geography, so targeting the United States and California means everyone in the United States. If you want individual regions, add only the regions.

The additive side, which is where the audience is actually built

Interests and follower look-alikes work as OR. Add more and the pool grows.

• Interest targeting runs on a fixed taxonomy. X states there are 82 interest categories, "fully aligned with the topics categorized on the platform", and that interests are inferred from who a person follows, what they repost, click on and post. So it is broad by design, and X's own advice is to choose interests that align directly with the creative rather than everything plausible.
• Follower look-alikes target people whose interests resemble the followers of accounts you name. X's example is entering a business account's handle to reach people likely to be interested in advertising on X. Three pieces of X's own guidance here are the most actionable targeting advice on its help centre. Aim for around 30 handles per campaign, with bulk upload available in the campaign form. Pick handles tied closely to your business rather than famous ones, because millions of followers are not millions of people aligned on your topic. And do not put interest targeting in the same campaign as handle targeting, because the two together widen the audience past the point where either can be read. Run broad interests as a separate campaign.
• Post engager targeting reaches people who already engaged with your posts, which is the cheapest warm audience on the platform and the one most accounts forget they have.
• Device, carrier and new mobile user targeting is a separate set, and the new mobile user option is the one app advertisers care about.

Custom Audiences, and the minimum that decides whether you can use them

X describes Custom Audiences as groups you create "for retargeting, exclusion, and/or expansion", built either from uploaded lists or from web activity. The list route matches your own customer or CRM data, such as email addresses, phone numbers or handles, against active users on the platform.

The number to plan around is the floor. X states that "to respect user privacy, the minimum size for a Custom Audience to be usable is 100 X users", and that an audience under that threshold displays as Audience too small and cannot be selected at all. Because an upload is matched rather than used whole, the list you upload has to be comfortably larger than 100 rows to clear 100 matched users, and how much larger depends entirely on your own data quality.

The exclusion use is the one that pays for itself fastest and gets skipped. Excluding existing customers from a prospecting campaign costs nothing and stops you buying reach you already own.

The default that quietly overrides your targeting

Optimized Targeting is on unless you turn it off. X's documentation says it "is pre-selected for all campaigns as a default" and sits in the Advanced targeting module within supported objective types. What it does is reach people beyond your targeted audience when X's model sees better performance available, using an engagement-based model built on on-platform and off-platform behaviour.

The important sentence is which parts it will not override. X states that with Optimized Targeting enabled, "certain targeting selections will always be respected (location, age, gender, etc) while others will be treated flexibly". So your demographic floor holds and your interest and handle work is advisory. That is a sensible trade on a performance campaign and a problem on a campaign whose whole point was a tight list, so decide deliberately rather than inheriting the default. Audience Expansion is a separate control in the same family and deserves the same decision.

How to read whether the targeting worked

X's Audience Insights, reached from the Analytics page, breaks a campaign's audience down by locations, keywords, handles, behaviours, gender, language, platform and interests, with an export at audience level. Two readings are worth taking every time.

• Compare the audience you got with the audience you asked for. If handles and interests ran in the same campaign, this is where you find out which one carried it, and the answer is usually interests because they are broader.
• Check the geography against your selections. If impressions appear outside the target, X's documented first check is the campaign History tab, because targeting changed after launch means the pre-change impressions went wherever the original setup allowed.

X also publishes an anti-discriminatory targeting policy and rules on targeting of sensitive categories. Read both before building audiences in housing, employment, credit, health or politics, where the available controls are deliberately narrower than the ad platform's general capability.

A build order that avoids the usual mess

• Set location, language and age deliberately, and leave gender unset unless the product genuinely is gendered. These are the floor and they are permanent.
• Upload your customer list as an exclusion first, before you build a single prospecting audience.
• Build three separate campaigns rather than one layered audience. One on competitor and industry media handles, one on broad interests, one on post engagers and site visitors.
• Decide Optimized Targeting per campaign. On for the broad performance campaign, off for the tight handle test you want a clean reading from.
• After a week, read Audience Insights and cut whatever the data says you did not mean to buy.

Questions about targeting on X Ads

How precise can geographic targeting get on X?

Countries, regions, metros, cities and postal codes are documented, with the stated caveat that some countries do not offer the granular options. Where a location does not appear, X advises targeting the closest region instead.

Why are my ads serving outside the countries I selected?

X's documented first check is the campaign History tab, which shows when targeting was changed. Adding geo targeting after launch means impressions before that change were served under the original settings.

How many accounts should I target with follower look-alikes?

X recommends around 30 handles per campaign, and recommends against running interest targeting in the same campaign, because the two together widen the audience past the point of being readable.

What is the minimum size for a Custom Audience?

100 X users. Below that, X shows Audience too small and the audience cannot be selected. Since an uploaded list is matched rather than used whole, upload comfortably more rows than that.

Should I leave Optimized Targeting on?

It is on by default and it will go beyond your interest and handle selections while respecting location, age and gender. Leave it on where performance is the goal, and turn it off where the reading from a specific audience is the goal.

Is interest targeting enough on its own?

It can carry a reach campaign, because the 82 categories are broad by design. For anything that needs precision, pair the creative with a handle-based or list-based campaign and keep the two apart so each one can be judged.

The audience no ad platform lets you target

Everything above is about reaching people while they scroll. A growing share of buyers now shortcut that entirely. They ask an assistant which supplier to use, get back two or three company names with a sentence each, and never see a feed. Those names are not an audience you can bid on, and no targeting control on any platform reaches them, because the selection happened inside an answer assembled from what has already been published about each company. It is worth knowing what that answer says in your category, and the cheapest way to find out is to type the question a customer would ask into two assistants, leaving your own company name out of the prompt so the result is not primed, and write down who came back.

We have not put our own product in this page, deliberately. We measure what assistants say about a business, and a reader who came here for X targeting controls is buying advertising operations, which is not a thing we sell.