> For the complete documentation index, see [llms.txt](https://docs.elevateab.com/elevate-helpcenter/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.elevateab.com/elevate-helpcenter/experiment-setup/audience-targeting.md).

# Audience Targeting

***

Audience targeting controls which visitors are eligible to enter an experiment. It is available on every experiment type, at **Step 4 of 5: Audiences**, and it works the same way whether you are testing a price, a page, a theme, an image, or custom code.

You can target as loosely as "mobile visitors" or as tightly as "logged out visitors from the UK market, arriving from a Meta campaign, landing on a specific product page". Thirteen dimensions are available and they combine, so the practical limit is your traffic rather than the tool.

Every rule you add makes your audience smaller and your experiment slower. That trade is the whole discipline of this page.

### Two ways to target

| Method                     | Where                                                  | Use it for                                                                                        |
| -------------------------- | ------------------------------------------------------ | ------------------------------------------------------------------------------------------------- |
| **Quick selectors**        | The **Audience Targeting** card on the Audiences step  | Device, visitor type, traffic source, country, and single UTM values                              |
| **Create Custom Audience** | The **Create Custom Audience** button on the same card | Everything else, multiple conditions, OR logic, and any dimension not exposed as a quick selector |

Quick selectors cover the common cases in one click. The custom audience builder is where the real narrowing happens.

### Quick selectors

The **Audience Targeting** card exposes four groups. Each defaults to including everyone, so an experiment with nothing changed here runs on all traffic.

#### Device Type

Select **All Devices**, **Desktop**, **Tablet**, or **Mobile**.

Device is the most common targeting decision for a reason. A change to a sticky bar, a drawer, or an image gallery often behaves completely differently across breakpoints, and a mobile win averaged against a desktop loss reads as no result at all.

#### Visitor Type

Select **All Visitors**, **New Visitors Only**, or **Returning Visitors Only**.

A visitor counts as returning when Elevate A/B Testing recognises them from a previous visit in the same browser. Someone who clears cookies, switches browsers, or moves from phone to desktop is treated as new.

This distinction matters more than it looks. New visitors have no prior expectation of your store, so they are the honest audience for a price change or a redesign. Returning visitors are the honest audience for anything about repeat purchase, and they will react to a price change partly because they remember the old price.

#### UTM Parameters

Four buttons, one per parameter: **Select utm\_source**, **Select utm\_medium**, **Select utm\_campaign**, and **Select utm\_content**.

Selecting one opens the **UTM Parameter** dialog:

1. Confirm or change the parameter in the **UTM Parameter** dropdown.
2. Choose a **Rule**.
3. Type a value in **Value** and press **Enter** to add it. Add as many values as you need.
4. Select **Apply & Confirm**.

UTM parameters are tags you add to your own links, in ads, emails, and social posts. If your links do not carry them, these rules match nothing.

#### Traffic

Two dropdowns, **Source** and **Country**, both defaulting to **All**.

**Source** is the referrer classification: direct, search, and the major social and ad platforms. **Country** covers all countries.

### Create Custom Audience

Select **Create Custom Audience** to open the **Add Custom Audience** dialog and build rules from the full dimension set.

Each rule is three fields:

| Field         | What it does                  |
| ------------- | ----------------------------- |
| **Dimension** | What you are matching on      |
| **Rule**      | How you are matching it       |
| **Value**     | What you are matching against |

Build one:

1. Select **Create Custom Audience**.
2. In **Rule #1**, choose a **Dimension**. The rule group takes the dimension's name once selected.
3. Choose a **Rule**. The available rules change with the dimension.
4. Enter or select a **Value**.
5. Select **Add another rule** for each further condition.
6. Set the group logic to **AND** or **OR** using the selector between rules.
7. Select **Save Rules**.

Use the delete icon beside a rule to remove it.

### Dimension reference

Thirteen dimensions are available in the custom audience builder.

| Dimension                   | What it matches                                   | Example value                           |
| --------------------------- | ------------------------------------------------- | --------------------------------------- |
| **Device Type**             | The device the visitor is browsing on             | Mobile                                  |
| **Visitor Type**            | Whether the visitor has been to your store before | New                                     |
| **Country**                 | The visitor's country                             | United States                           |
| **Market**                  | The Shopify market the visitor is assigned to     | Europe                                  |
| **Traffic Source**          | Where the visitor came from                       | Direct, Google, Facebook                |
| **Page URL**                | Any part of the URL the visitor is viewing        | `/products?id=123` or `/products`       |
| **Referring Domain**        | The domain the visitor arrived from               | `facebook.com`                          |
| **Entry Page**              | The first page URL the visitor landed on          | `/products/shoes`                       |
| **utm\_source**             | The ad platform or website                        | `facebook`, `tiktok`, `google`          |
| **utm\_medium**             | The type of ad platform                           | `social`, `paidsocial`, `search`, `cpc` |
| **utm\_campaign**           | The specific campaign                             | `black-friday-sale`                     |
| **utm\_content**            | The specific ad or creative                       | `black-friday-video`                    |
| **Customer Authentication** | Whether the visitor is logged in                  | Logged in                               |

Three of these are worth knowing about specifically, because they unlock targeting the quick selectors cannot reach.

#### Market

**Market** targets Shopify markets rather than countries. If you run markets with their own currencies, pricing, or catalogues, this is the dimension that matches how your store is actually configured. Targeting Country when your pricing is set per market gives you a segment that does not correspond to a pricing rule, which is the wrong shape for a price experiment.

#### Referring Domain

**Referring Domain** matches the exact domain the visitor came from, where **Traffic Source** matches a broad classification. Use Traffic Source for "all social traffic" and Referring Domain when you need one publisher, one affiliate, or one partner site separated from the rest.

#### Customer Authentication

**Customer Authentication** separates logged in visitors from logged out ones. Logged in visitors are typically existing customers with saved details and a shorter path to purchase, so mixing them with first time logged out visitors averages two very different funnels into one number.

This is the dimension to reach for when testing anything account related: a returning customer promotion, a saved cart experience, a members only price, or a checkout that behaves differently for a recognised customer.

### Rules

The rules available depend on the dimension. Text dimensions such as Page URL and the UTM parameters offer the full set. List dimensions such as Country and Device Type offer membership rules.

| Rule                 | How it works                                         |
| -------------------- | ---------------------------------------------------- |
| **Is one of**        | Matches if the value is any of the selected options  |
| **Is not**           | Matches if the value is none of the selected options |
| **Equals**           | Matches only an exact match of the text you enter    |
| **Not Equal**        | Matches anything other than the text you enter       |
| **Contains**         | Matches if the text appears anywhere in the value    |
| **Does not contain** | Matches if the text does not appear in the value     |
| **Starts with**      | Matches if the value begins with the text            |
| **Ends with**        | Matches if the value ends with the text              |

Two rules that catch people out:

**Equals on a URL.** Page URL matches against any part of the URL including query parameters, so **Equals** on a bare path fails for every visitor arriving with tracking parameters attached. Use **Contains** or **Starts with** for anything receiving campaign traffic.

**Equals on a UTM value.** UTM values match what your links actually carry. If half your ads send `Facebook` and half send `facebook`, **Equals** catches one half. Use **Contains**, or fix the links.

### AND and OR logic

The selector between rules sets the logic for the group. With **AND**, the dialog reads **All conditions in this group must match**, and a visitor must satisfy every rule. With **OR**, any single rule is enough.

The distinction is the difference between narrowing and widening:

* **AND** narrows. Each rule removes visitors. Three ANDed rules can easily leave you with two percent of your traffic.
* **OR** widens. Each rule adds visitors. Use it to gather several sources or markets into one audience.

Before reaching for OR on a single dimension, check whether the rule already takes multiple values. **Country is one of** accepts a list, and **Value** in the UTM dialog accepts several values pressed in one at a time. One rule with three values is simpler to read later than three ORed rules.

### Worked examples

#### Mobile shoppers in your European market

```
Device Type      Is one of    Mobile
AND
Market           Is one of    Europe
```

For a layout change that only exists on mobile, in the market where your pricing and catalogue are set up for it.

#### Cold traffic from one paid campaign

```
Customer Authentication   Equals       Not logged in
AND
utm_medium                Equals       paidsocial
AND
utm_campaign              Equals       spring-launch
```

The audience a landing page built for that campaign was actually designed for. Existing customers who happen to click the ad are excluded, so they do not flatten the result.

#### One affiliate partner's traffic

```
Referring Domain    Equals    partnersite.com
AND
Entry Page          Contains  /collections/
```

Isolates a single partner's referrals arriving on collection pages, which is the setup for testing a landing experience negotiated with that partner.

#### Returning customers, several markets

```
Customer Authentication   Equals       Logged in
AND
Market                    Is one of    UK, Europe
```

A returning customer experience across two markets, without pulling in first time visitors whose behaviour would tell you nothing about repeat purchase.

### How narrow is too narrow?

Every AND multiplies the reduction. Take the cold paid traffic example: if mobile is 60 percent of traffic, paid social is 15 percent of that, and this campaign is a third of paid social, you are running on roughly 3 percent of your visitors.

On 20,000 visitors a week, that is 600. An experiment needing 10,000 per arm will not finish this year.

Do the arithmetic before launching, not after two silent weeks. If the number is too small, you have three options:

| Option                                | When it works                                              |
| ------------------------------------- | ---------------------------------------------------------- |
| Drop the least essential rule         | The rule was preference rather than hypothesis             |
| Launch broadly and segment the report | You want the insight, not the isolation                    |
| Accept a longer run                   | The traffic supports it and the question is worth the wait |

The general rule: target for the hypothesis, not for preference. "Does this work on mobile?" earns a device rule. "I mostly care about mobile" does not, because the report can tell you that afterwards without costing you the sample.

### Targeting, isolation, and allocation

Three controls decide who sees what. They are separate, and they apply in this order:

| Control                | Question it answers                                                        |
| ---------------------- | -------------------------------------------------------------------------- |
| **Audience targeting** | Is this visitor eligible for the experiment?                               |
| **Traffic isolation**  | Of the eligible visitors, what share should enter?                         |
| **Traffic allocation** | Of the visitors who entered, who sees control and who sees each variation? |

Targeting filters first. Isolation then takes its percentage of whoever survived the filter, not of your total traffic. A 10 percent isolation on an audience that is already 3 percent of your store leaves you with 0.3 percent.

See [Traffic Isolation](https://claude.ai/elevate-helpcenter/operate/experiment-setup/traffic-isolation) for the second control, which is also Elevate A/B Testing's mutual exclusivity mechanism.

### When targeting is evaluated

Targeting is checked when a visitor first becomes eligible. If they match, they enter and are assigned a variation, and that assignment persists across sessions. If they do not match, they see your original experience and never appear in the experiment's results.

This means a visitor whose circumstances change later does not move. Someone who first arrived on desktop stays in the desktop assignment even if they return on mobile, because the experiment recognises them as an existing participant rather than re-evaluating them.

### Targeting versus personalizations

Targeting inside an experiment decides who is measured, against a control, for a limited period. A [Personalization](https://claude.ai/elevate-helpcenter/personalizations/overview) decides who permanently receives an experience, with no control group and no measurement of lift.

The productive sequence runs one into the other: launch an experiment broadly, read the segmented report, find the segment that responded, then build a personalization for that segment. Targeting narrowly from the start skips the step that tells you which segment was worth targeting.

### Next steps

* [Traffic Isolation](https://claude.ai/elevate-helpcenter/operate/experiment-setup/traffic-isolation) for capping what share of your eligible audience enters the experiment
* [Experiment Lifecycle](https://claude.ai/elevate-helpcenter/operate/experiment-setup/experiment-lifecycle) for when targeting is evaluated and when a visitor's assignment locks in
* [Reading Your Results](https://claude.ai/elevate-helpcenter/analyze/reports/reading-your-results) for segmenting a broadly launched experiment instead of targeting it narrowly
* [Personalizations](https://claude.ai/elevate-helpcenter/personalizations/overview) for permanently serving an experience to a segment once an experiment has proven it