Custom Segments in Google Ads 2026: Complete Guide

Google’s off-the-shelf audiences — affinity, in-market — are convenient, but they describe your market in someone else’s vocabulary. A category like “Business Software” lumps together the person shopping for a $20/month CRM and the person evaluating a $200,000 ERP. Custom segments in Google Ads exist to fix exactly that: you describe the audience with your own inputs — search terms, websites and apps — and Google finds people whose behaviour resembles them.

This guide covers how to build a segment that actually lands on your buyer, where segments target directly versus where they only act as a hint to the algorithm, and which construction mistakes reliably burn budget.

What a custom segment is and how it differs from stock audiences

A custom segment is an audience you define; Google then finds people with comparable interests and behaviour. The output is a modelled group, not a list of specific individuals.

Audience type	Who defines it	When to use it
Affinity	Google, based on long-term interests	Top of funnel, broad reach, awareness
In-market	Google, based on active purchase signals	Mid funnel, when the stock category matches your market
Custom segments	You, via search terms, URLs and apps	When no stock category describes your niche
Your own data (remarketing, Customer Match)	You, via real contacts and events	Bottom of funnel, working an existing base

Custom segments close the gap between “far too broad” and “only people who already visited us”. They are the only way to describe a niche Google has no label for: people researching warehouse automation for marketplace sellers, or people who visit four specific competitor sites.

Three input types and what each one actually does

Keywords and phrases

You list interests as words. There is an important fork here: you can enter general interests, or you can specify that these are search terms your ideal customer uses on Google properties. The second option is sharper on intent but limits delivery to Google properties. That is the reach-versus-precision decision, and it deserves a deliberate choice rather than a default.

Website URLs

You list sites your audience visits. The most misunderstood point in the whole feature: this is not placement targeting. Google will not run your ad on a competitor’s site. It finds people whose interests resemble those of that site’s visitors and shows them ads elsewhere. If you want ads on specific sites, that is managed placements — a different mechanism entirely.

Apps

You list apps your audience uses. Same logic: your ads reach people who download and use similar apps, rather than appearing inside the apps you named.

All three input types describe a person, not a place. A custom segment answers “who is this buyer”, never “where should the ad appear”.

Where custom segments in Google Ads work — and where they do not

Direct targeting: Display, Gmail, Demand Gen and Video campaigns. Here the segment is applied as targeting or as observation.
Not applied directly: Performance Max. You cannot attach a custom segment as targeting, but you can use it as an audience signal. The distinction matters: a signal does not restrict delivery, it accelerates learning by telling the algorithm where to start.
Search campaigns are built around keywords; audiences there are used primarily in observation mode as a data and bid-adjustment layer rather than as primary targeting. That mode is covered in our guide to audience observation in Google Ads.

The practical implication for 2026: in manually targeted campaigns a segment is a filter; in automated ones it is a hint. If you expect an audience signal in PMax to hard-cap delivery, you will be disappointed — the algorithm starts with your segment and moves beyond it the moment it finds conversions elsewhere. The mechanics are unpacked in the complete Performance Max guide.

Building a segment that lands

Step 1. One segment, one intent

The most common failure is one catch-all segment: problem-awareness terms, comparison terms, pricing terms and competitor URLs all in the same bucket. The result is a blurred audience with no clear intent — and no ad copy can speak to it.

Split by stage and intent instead:

Problem awareness: “how to automate order tracking”, “why we keep losing leads”.
Solution evaluation: “CRM comparison”, “best CRM for retail”, “CRM reviews”.
Vendor selection: competitor sites, competitor brand terms, directories and review aggregators.
Adjacent consumption: apps and services your buyer uses alongside your category.

Step 2. Pick inputs that are unambiguous

A good input is one your audience uses and outsiders do not. A bad input is a generic word with five meanings. “Rental” is a bad input; “excavator rental daily rate” is a good one. The same applies to URLs: a narrow specialist competitor produces a precise audience, while a giant marketplace produces roughly the entire internet.

Step 3. Give the model enough material

Google is building a model, and models need examples. A workable rule of thumb is 10–15 homogeneous inputs per segment. Three gives too little signal; fifty mixed ones dissolve the meaning. Every input inside one segment should describe the same person.

Step 4. Sanity-check reach before launch

The builder shows a reach estimate. Zero or near-zero means the segment will never accumulate volume and the campaign will either stall or expand on its own. An enormous estimate means your inputs are too generic and you have rebuilt an in-market category by hand.

Step 5. Observe before you target

Where the campaign type allows it, add the segment in observation mode for a week or two and watch CTR, conversion rate and CPA against the rest of the traffic. It is the cheapest possible test of the hypothesis, and it costs you no reach.

Custom segments in Demand Gen and on YouTube

Demand Gen is the primary use case today. The format lives on visual surfaces where there are no keywords, so audience quality carries nearly all the weight.

Search-term segments effectively port search intent into a feed environment: the person searched on Google and sees your ad on YouTube or Discover. It is the closest thing to performance targeting available at the top of the funnel.
Competitor URL segments for conquesting. They work well when competitors are narrow and thematic, poorly when they are general-purpose portals.
Combination with your own data. Custom segments bring new people; Customer Match and first-party data exclude existing buyers and seed lookalike modelling. Keep acquisition and retention in separate campaigns so they do not compete for the same budget.

Format specifics live in our guides to Demand Gen campaigns and YouTube and Shorts video ads. One rule holds across both: the sharper the segment, the more specific the creative has to be. An audience assembled from “CRM comparison” searches expects a comparison, not a brand film.

Using segments as Performance Max signals

Make the signal narrow. A broad signal tells the algorithm nothing it would not have done anyway. Value comes from a segment you could never have assembled from stock categories.
Combine signal types. A search-term custom segment plus your customer list are two different hints, and together they outperform either alone.
Do not expect containment. If you need hard control over who sees the ad, PMax is the wrong tool — use Demand Gen or Display with direct targeting.
Judge it over weeks, not days. Changing a signal changes the learning inputs, so effects take time. Record the date so you can later reconcile it in change history against the performance curve.

Privacy in 2026

Custom segments are built on behavioural modelling rather than lists of personal data, so they have weathered privacy tightening better than classic cookie remarketing. Two constraints still apply:

Consent affects data volume. A misconfigured consent setup shrinks both reach and model accuracy. The mechanics are covered in our piece on Consent Mode v2 and ad privacy.
Sensitive categories are off limits. You cannot build segments around health, financial status, religion or other sensitive attributes — such inputs fail policy review. This is not a formality: attempts to route around it get audiences rejected and put the account under scrutiny.

The durable hedge is your own data, ingested correctly — see Data Manager and first-party data.

Four ready-made segment blueprints

Abstract advice transfers badly to a real account, so here are four skeletons you only need to fill with your own inputs.

Mid-market B2B SaaS

“Evaluation” segment: terms like “[category] comparison”, “[category] for [industry]”, “[competitor] alternative”, “[category] reviews”. Twelve to fifteen inputs, every one of them about active evaluation.
“Competitors” segment: URLs of five to eight narrow, category-specific competitors plus two or three industry directories. Leave the giant portals out.
“Adjacent stack” segment: apps and services the same buyer uses — task trackers, accounting, analytics.
Exclusion: your current customer list via first-party data.

Niche e-commerce

“Purchase intent” segment: queries containing models, specifications and words like buy, price, delivery.
“Comparison” segment: review and ranking queries plus URLs of specialist review sites.
“Season / occasion” segment: built for a specific date and switched off afterwards, so you are not dragging irrelevant reach around all year.

Local business and services

“Problem” segment: queries describing the customer’s situation rather than the name of the service — people search the symptom far more often than the term.
“Choosing a provider” segment: queries with “near me”, “cheap”, “reviews”, plus aggregator and directory URLs.
Always check the interaction with geo settings: a narrow segment plus a narrow radius frequently produces zero reach.

App or subscription product

App-based segment: direct analogues and companion products. This is the most underused input type of the three.
Query-based segment: “app for [task]”, “how to [task] on phone”.
Exclusion: active subscribers — otherwise acquisition and retention bid against each other for the same impression.

How to measure whether a segment works

A segment is a hypothesis and it should be closed with data, not impressions. Four metrics are enough:

Metric	What it tells you	What to do if it is poor
CTR versus campaign average	Whether the message lands on this audience	Change the creative, not the segment — the audience may be right
Click-to-conversion rate	Whether the segment’s intent matches your offer	Rebuild the inputs around a different funnel stage
CPA / cost per target action	Whether the segment is economically viable	Narrow the inputs or reduce its budget share
Share of new users	Whether it brings new people or overlaps remarketing	Add exclusions for existing customers

One important caveat: do not judge a segment on its first three days. Audience-driven campaigns stabilise more slowly than Search, and any change to the input list effectively rebuilds the model from scratch. A sensible evaluation window is 10–14 days at sufficient impression volume.

Common mistakes

Treating URLs as placements. A competitor-URL segment does not put your ad on the competitor’s site. It describes a person, not a location.
Mixing funnel stages in one segment. “What is a CRM” and “buy CRM for 50 seats” are two different people needing two different ads.
Three inputs per segment. Too little material for the model; reach lands near zero.
Generic single words. They convert your segment into a broad category — without the data quality of a real one.
Expecting a PMax signal to restrict delivery. A signal is a starting hint, not a filter.
One creative for every segment. A precise audience with a generic message throws away the precision you just paid for.
Not excluding existing customers. In acquisition campaigns, exclude buyers — otherwise you pay to reach people who already converted.

FAQ

What is a custom segment in Google Ads?

An audience you define yourself using keywords and phrases, website URLs and app names. Google then reaches people with similar interests and behaviour. It is a modelled group, not a list of identified users.

Which campaign types support custom segments?

Directly: Display, Gmail, Demand Gen and Video campaigns. Performance Max does not accept them as targeting, but they can be applied as an audience signal.

Will my ads appear on the websites I enter as URLs?

No. Those URLs describe audience interests, not placements. To appear on specific sites you need managed placements, which is a separate mechanism.

How many inputs does a segment need?

A practical guideline is 10–15 homogeneous entries. Fewer starves the model of signal; many more, and mixed, dissolves the segment into a generic category.

What is the difference between search-term and interest inputs?

Choosing the search-term option targets based on what people searched for on Google properties and limits delivery to those properties. It is sharper on intent but narrower in reach than a segment built as general interests.

Can I use custom segments in Search campaigns?

Search is built around keywords, and audiences there are used mainly in observation mode as an analytics and bid-adjustment layer. The primary home for custom segments is Demand Gen, Video and Display.

Does a custom segment work as a Performance Max signal?

Yes, but strictly as a learning hint rather than a delivery cap. The algorithm starts with your segment and expands beyond it once it finds conversions elsewhere.

How do I know a segment is well built?

Three signs: the reach estimate is neither zero nor enormous; in observation mode it beats campaign-average CTR and CPA; and you can write a specific ad for it without falling back on generic language.

Can I target competitors’ customers?

You can build segments from competitor URLs and competitor brand searches — that is permitted. What is not permitted is using third-party trademarks in ad copy where policy forbids it, or building segments around sensitive categories.

Does Consent Mode affect custom segments?

Indirectly, yes. The less data available because of refusals or a misconfigured consent setup, the smaller and less accurate the modelled audiences. The effect shows up first in European traffic.

How often should segments be refreshed?

Quarterly is a sensible cadence: search language shifts, competitors come and go, apps lose relevance. Add an off-cycle review whenever you launch a new product or enter a new market.

My segment gets no impressions. What should I check?

In order: whether there are enough inputs; whether they are too narrow; whether other restrictions are layered on top (geo, devices, schedule, a second audience); and whether budget and bids are competitive. The cause is usually stacked narrowing conditions rather than the segment itself.

The short version

Custom segments let you describe an audience in your own vocabulary where Google’s stock categories are too coarse. All three input types — search terms, URLs, apps — describe a person, never a placement. In Demand Gen, Video and Display they act as targeting; in Performance Max they act as a learning signal. Build one intent per segment, feed the model 10–15 homogeneous inputs, check reach before launch, and write dedicated creative for each segment.

Related reading: bid, budget and target simulators for when the segment works and it is time to model scaling, and Google Ads change history for knowing precisely which edit caused what. Infrastructure for stable launches sits under agency Google Ads accounts; the rest of the library is in the PPC Rebels blog.