Why you have to clean up your campaign names before AI analyzes your ad accounts

AI reads names before it reads numbers

Say you ask AI: “Find the campaigns whose ROAS by product group has dropped over the last 14 days.” If your campaign names look like new_test_2, 전환_0501, meta_리타겟, 혜택소재_수정본, SA_브랜드_복사본, AI has no way to tell which product group or which funnel stage a campaign belongs to.

The answer usually goes one of two ways: it plays it so safe that you get a useless summary, or it confidently spits out the wrong classification. The problem isn’t that AI is dumb. It’s that the account isn’t structured in a language AI can read.

A campaign name isn’t housekeeping—it’s a data contract

Campaign names aren’t labels you slap on to make things look tidy for humans. They’re a data contract that lets your ad platforms, GA4, back office, spreadsheets, and AI agents all refer to the same thing in the same way.

When that contract breaks, AI loses the context even when it has the numbers. From a name like 전환_0501, it can tell the campaign has a conversion objective, but not which product group, which funnel stage, or which test it belongs to.

GA4, too, breaks out source, medium, and campaign from your UTM values. Google recommends using utm_source, utm_medium, and utm_campaign consistently, and notes that values with different capitalization are treated as different. SpringSale and spring_sale may look like the same campaign to a person, but they can show up as separate rows in your reports.

How bad naming wrecks AI’s answers

In AI analysis, classification errors are more dangerous than calculation errors. ROAS can be calculated correctly, but if the campaigns are bucketed wrong, the conclusion is wrong. A claim like “ROAS on skincare prospecting campaigns is down” might actually be a blend of retargeting and brand search campaigns.

The minimum AI needs to read a campaign

You don’t need to cram every detail into the campaign name. When names get too long, no one follows the convention. Instead, define the minimum set of fields AI needs to run analysis.

For example, meta_purchase_prospecting_skincare_broad_202605 is something AI can read. Whether you go with Korean or English is the team’s call. What matters is that you don’t mix forms like 스킨케어, skincare, and skin—pick one and stick with it.

When names aren’t enough, lean on UTMs and a mapping table

Try to pack every bit of context into the campaign name and you’ll fail. Each channel has different limits on name length, structure, and auto-parameter support, and the moment you rename a live campaign you can break the continuity of your reports. It’s safer to apply your rules to new campaigns and fix historical ones with a mapping table.

UTMs are how GA4 splits your traffic; the mapping table is how your team interprets campaigns. The tracking-parameter tooling differs by channel—Google Ads URL options, for one—but in the end GA4 and Neuro need to read consistent source, medium, campaign, and product-group values.

In Neuro, you capture this kind of mapping standard as a table in your workspace guidelines or in a naming-audit Gem. If the rules get complex, you can keep them as JSON instead. The point is to stop making AI guess “Which product group does this campaign belong to?” every single time.

In Neuro, start with a naming audit

Before you ask AI to run performance analysis, have it audit the account’s classification rules. The goal isn’t to interpret performance—it’s to confirm AI can read the account reliably in the first place.

The key line here is “Don’t guess—flag them as unclassifiable.” A good AI answer isn’t one that fills the blanks with plausible guesses. It clearly marks what it doesn’t know and tells the team which rules to fix.

How to use this in Neuro

1. Using your connected ad accounts and GA4, pull the active campaigns from the last 90 days. Instead of jumping straight to performance analysis, have AI classify them from campaign names and UTMs first.
2. Sort the campaigns AI flagged as unclassifiable by budget or revenue share. What matters isn’t the total campaign count—it’s whether your high-spend campaigns can be read.
3. Fix new campaigns to match your naming rules; correct historical ones with product-group/funnel mapping tables. Don’t go on an aggressive bulk-rename spree across live campaigns.
4. Save the finalized rules in your workspace guidelines or a Neuro Gem. For example: “Group 스킨케어, skincare, and skin all under 스킨케어”—so the team’s interpretation rules are written down.
5. Check the unclassifiable rate once per new-campaign setup cycle and once a quarter. If your team launches a lot of new campaigns, make it a monthly routine.

Prompts you can use today

What these prompts share is that they don’t ask AI for the conclusion up front. They first check whether the account is readable, separate out the uncertain items, and only then move on to performance analysis. If you read this far, you should be able to paste them straight into Neuro.

Minimum rules your team must define

• Use the same name for the same product group.
• Always separate prospecting, retargeting, and brand search.
• Leave a test ID on every test campaign.
• Ban names that contain only new, test, copy, or 수정본.
• Standardize UTM values to lowercase with a single delimiter rule.
• Have AI flag items as unclassifiable instead of guessing when it doesn’t know.
• Fix historical campaigns that are hard to rename with mapping tables.

AI ad analysis starts with reading the account. Before you hand your ad accounts to AI in Neuro, the first question isn’t “Why did performance drop?” It’s “Can AI actually read this account correctly?”