How UTMs are read
When a visitor arrives, GA4Google's event-based analytics platform. looks at the landing URL for the parameters it documents:
utm_source, utm_medium, utm_campaign, utm_term, utm_content, plus the newer utm_id,
utm_source_platform, utm_creative_format and utm_marketing_tactic. Their values populate the
corresponding traffic-source dimensions for the session and, under the property's attribution settings,
for the user and the conversion. Where no parameters are present, the tool falls back to the referrer:
a known search engine becomes organic search, a known social site becomes organic social, another website
becomes referral, and no referrer at all becomes direct.
The channel reports do not show raw parameters; they show channel groups derived from them. Google documents the default channel group definitions on its help pages: each channel is a rule over source, medium and sometimes campaign name. Paid search, for example, requires a medium matching the documented paid pattern and a source on the documented list of search sites; email requires a medium matching the documented email variants; organic social requires a source on the social list with a non-paid medium. A visit whose values match no rule lands in "unassigned", which is where badly tagged traffic goes to be forgotten. The reports and dimensions guide shows where these dimensions appear and how custom channel groups can reinterpret them; the point here is that the rules read the medium above all, so the medium is where consistency matters most.
Values are read exactly as written. Google does not merge case variants or separators, so a medium written
as Email in one link and email in another produces two values, and if the channel rule matches one case
and not the other, two channels. Everything in the convention below follows from that single fact.
The five parameters and what each should carry
Each parameter answers one question, and the convention assigns a vocabulary to each.
| Parameter | Question it answers | What it should carry | What it should not carry |
|---|---|---|---|
utm_source | Where did the click originate? | The platform or property: google, meta, linkedin, newsletter, a partner's domain | The medium ("paid"), the campaign, free text |
utm_medium | What kind of channel was it? | A short controlled list aligned to the channel rules: cpc, paid_social, email, affiliate, display, referral | Platform names, campaign names, anything not on the list |
utm_campaign | Which initiative paid for or planned the link? | A structured name from the campaign convention: market, objective, theme, period | Spaces, free text, the ad's headline |
utm_content | Which variant within the campaign? | Creative, placement or link position: hero_banner, footer_link, video_a | Anything already in source, medium or campaign |
utm_term | Which keyword or audience? | The paid keyword, or an audience name where keywords do not apply | Free text; nothing at all in auto-tagged search |
The medium list is the heart of the convention. Keep it short, publish it, and map every value to the channel it should land in so that anyone can check a link against the channel report. A new medium is a governance decision, not something a campaign manager invents on a Friday.
utm_id deserves a note: it is documented as a campaign identifier that can join cost data imported into
analytics to the sessions the campaign produced. If cost import is used, the convention assigns ids; if not,
it can be left empty without consequence.
Case, separators and forbidden characters
Three rules, applied without exception:
- Lower case only. Every value, every parameter, every link. Mixed case is the single largest source of duplicate rows in channel and campaign reports.
- One separator inside a value, a different one between fields. Underscores within a field
(
paid_social,hero_banner) and hyphens between the fields of a structured campaign name (uk-spring-prospecting), or the reverse — the choice matters less than its consistency, and the choice should match the one used in the tracking plan's event names so readers meet one style across the data. - No spaces, no reserved characters. A space becomes an encoded sequence that differs by the tool that encoded it; ampersands, question marks, hashes, equals signs and plus signs are URL syntax and break the link or truncate the value; non-ASCII characters are encoded unpredictably. Letters, digits, underscores and hyphens only.
Two further habits prevent subtler problems. Place the parameters after the page's own query string and before any fragment, so that the page still routes correctly and the parameters survive. And never let a redirect strip them: shortened links, vanity domains and tracking redirects must pass the parameters through to the final URL, which is a thing to test rather than assume.
Radio, podcast and print adverts depend on exactly such redirects, since a short vanity address is the only way a listener arrives tagged, and the guide to measuring offline media explains why those tagged visits are a floor rather than the campaign's whole response. Printed cards in an online store's parcels work the same way, and the guide to packaging and order inserts tags each insert version so that it reports under its own name.
A convention that survives growth
A convention survives when it is written down, when every field has a controlled vocabulary, and when nobody needs to remember it because a builder enforces it. The structure most teams converge on:
| Field | Where it lives | Vocabulary | Example value |
|---|---|---|---|
| Platform | utm_source | Fixed list: google, meta, linkedin, tiktok, newsletter, partner-name | meta |
| Channel type | utm_medium | Fixed list aligned to channel rules | paid_social |
| Market | First field of utm_campaign | Country or language codes from a fixed list | uk |
| Objective | Second field of utm_campaign | Short list matching the business objectives: prospecting, retargeting, retention | prospecting |
| Theme | Third field of utm_campaign | The initiative or product line, from the campaign brief | spring-launch |
| Period | Fourth field of utm_campaign | A quarter or month token in a fixed format | a quarter token |
| Variant | utm_content | Creative or placement, structured: format then version | video_a |
| Keyword or audience | utm_term | Keyword for manual paid search; audience name for social | lookalike_purchasers |
The campaign name is the field that carries the most information and is therefore the most fragile. Fix the order of its fields, the separator between them and the vocabulary of each, and the name becomes parseable: a report can split it by the separator and group by market or objective without anyone retyping anything. The campaign brief is where the name is assigned, before any link is built, so that every platform, every link and every report uses the same string.
Where ad platforms have their own campaign names — and they do — align the two. The Google Ads naming conventions guide covers the account-side convention in detail; the UTM campaign value should either equal the platform campaign name or be derived from it by a documented rule. A reader moving from a platform report to an analytics report should recognise the same campaign instantly. The same string belongs on the CRM side too, as the name of the campaign record in the CRM that lists the people the campaign reached. When the CRM's source report shows a large blank or ungrouped segment, the diagnosis for a blank or wrong CRM lead source traces it to the forms, imports and links that produced it.
One campaign, three links, one row in the report
A furniture retailer launches a spring sofa campaign in the UK with paid social, a newsletter and a partner
placement. The convention produces 3 links. Meta: utm_source=meta&utm_medium=paid_social&utm_campaign=uk-prospecting-spring-sofas-2026q2&utm_content=video_a&utm_term=lookalike_purchasers.
Newsletter: utm_source=newsletter&utm_medium=email&utm_campaign=uk-retention-spring-sofas-2026q2&utm_content=hero_banner.
Partner: utm_source=homestyle-mag&utm_medium=affiliate&utm_campaign=uk-prospecting-spring-sofas-2026q2&utm_content=article_link.
In the channel report the 3 links land in paid social, email and affiliate respectively; in the campaign
report, splitting the campaign value on hyphens groups 2 of the 3 under the prospecting objective and all 3
under the spring-sofas theme. Before the convention, the same campaign had appeared under 7 campaign values
across 2 months, including Spring Sofas, spring_sofas_FB and sofas-spring-2026, and the reporting
analyst had spent roughly 2 hours a week merging them by hand.
For affiliate programmes, the place to apply the convention is the link each partner receives on joining, with the values built in rather than typed by the partner, as the tutorial on onboarding and activating new affiliates sets out.
Auto-tagging versus manual tagging
Google Ads can append a click identifier to every ad link automatically, and when the Google Ads account is linked to the analytics property, the property uses that identifier to retrieve campaign, ad group, keyword and creative detail without any UTM parameters. This is documented behaviour and it is more complete than manual tags, because it carries fields a UTM cannot. For a linked property, the convention should say that Google Ads links are auto-tagged and that manual parameters are not added to them; where both are present, Google documents a property-level setting that decides whether manual values override the auto-tagged ones, and the convention should record which way it is set.
Other platforms differ. Meta appends its own click identifier, but analytics does not read campaign detail from it; a Meta ad link without UTMs is recorded from the referrer and lands in organic social rather than paid social, which is one of the commonest reasons paid social looks underreported in analytics. The same applies to most social and display platforms. For those, manual tagging with the convention is the only option, and most of them offer URL-parameter templates that insert campaign and ad names automatically — use them, with the convention's vocabulary, so that the values are generated rather than typed.
The decision rule is simple: auto-tag where the platform and the analytics tool share an identifier; manually tag everywhere else; never do both on the same link without a documented reason. The comparison of UTM parameters and auto-tagging works through the choice source by source and sets out the checks to make when there is a reason to use both.
Governance and a builder
A convention decays when more people build links than know the rules. Governance is the set of habits that stops that, and it is light:
- One document. The convention, with every vocabulary list, lives in one place that the campaign brief links to. Changes are made there and announced.
- One builder. A shared spreadsheet or small internal tool with dropdowns for every controlled field, a generated campaign name and a generated URL. Nobody types a parameter by hand. The builder also keeps a log of every link generated, which becomes the register for audits.
- One owner. Marketing operations, usually, who approves new vocabulary values, reviews the campaign report monthly for values that broke the rules, and traces each back to its source.
- A check before launch. The campaign brief's launch checklist includes "links generated from the builder and tested end to end", and a reviewer opens one link per platform to confirm the parameters survive redirects.
- A monthly report of exceptions. Sessions with a medium outside the list, campaign values that do not parse, unassigned channel traffic above its usual share. Each exception is a training conversation, not a blame exercise.
Governance here is cheaper than anywhere else in measurement because the failure is so visible: a campaign value that does not parse shows up in the first report. The measurement planning hub places naming as the fourth artefact in the sequence, after the tracking plan and before the quality routine, because the quality routine's job is partly to enforce what naming defined. Attribution (AttributionAssigning conversion credit to touchpoints.) depends on it entirely: a model can only credit channels it can identify, and every KPIA metric tied directly to an objective. cut by channel inherits the convention's discipline or its absence.
To start from a ready-made structure rather than a blank sheet, the UTM and naming conventions template provides fields and allowed values you can adapt to your own channels.
To keep the convention applied in practice, the UTM link request SOP describes how team members request tagged links so every campaign URL follows the agreed format.
When untagged links and stripped redirects have already inflated the Direct channel, the diagnosis page for direct traffic that is too high locates each leak by landing page.
If the convention exists but teams, platforms and agencies still produce different values, the diagnosis of inconsistent campaign names and UTM tags covers the operational causes behind the drift.
Tagged addresses are also extra URLs for search engines to crawl, which is why the URL parameter audit for SEO keeps tracking parameters off internal links and points tagged URLs at their clean canonical.
A local business's Google Business Profile carries links that arrive untagged unless someone tags them, and the tutorial on measuring Business Profile results applies the convention to the profile's website and booking links and checks where those visits are grouped.
Common mistakes
- Mixed case.
Email,emailandEMAILare three mediums. Lower case, enforced by the builder. - Inventing mediums. A medium not on the controlled list lands in unassigned; new values are a governance decision.
- Putting the platform in the medium.
utm_medium=facebookis a source value in the wrong field and breaks the channel rule. - Free-text campaign names. Names that cannot be parsed cannot be grouped; fix the field order and separator.
- Tagging internal links. A UTM on a link within the site starts a new traffic source mid-visit and misattributes the session.
- Trusting redirects. Shorteners and vanity URLs that strip parameters report as direct; test every link once.
- Manual tags on auto-tagged links without a reason. Two sources of truth for one click, and a property setting deciding silently which wins.
