Skip to main content

How the X-ray decides

What the X-ray compares each item with, when a difference becomes a problem, what it does with little data, what it reads on pages and creatives, and what it never does.

Examples use “Clients”. If you manage ads for your own company, this section is called “Companies”. The steps are the same.

5 minutes read
Updated Oct 31, 2026
Reviewed Oct 31, 2026

Copy the article as Markdown to paste it into an AI agent.

Before you begin

  • A client with at least one linked Meta Ads or Google Ads account.

Expected result

You will understand how the X-ray reaches each card: what it compares each item with, when a difference becomes a problem, what it does with little data, what it reads on pages and creatives, and what it never does.

The X-ray reads a client's campaigns, ad sets and ads in Meta Ads and Google Ads and shows what to change first. To do that, it puts together data that Ads Manager and Google Ads show on separate screens, such as audiences, search terms and results, and it reads the landing pages, which sit outside both.

It compares what pursues the same result

Each campaign, ad set and ad is compared with the others in the same account that pursue the same kind of result, and with the previous period. A lead is compared with a lead and a purchase with a purchase, from the same event.

Results of different kinds are never added together. A lead does not count as a purchase, and a video view does not count as a lead.

It tests before calling something a problem

A cost difference can be luck, especially with few results. Before pointing out a problem or a winner, the X-ray works out the chance that the difference is luck. Because an account runs many comparisons at once, this calculation accounts for how many there are, so a lucky difference does not become a card.

Each card shows the Confidence: High, Medium or Low. It says how much the data of the period supports that conclusion.

It waits for data

  • With many results in the period, the X-ray shows high-confidence findings first and fills in with medium-confidence ones.
  • With few results, the X-ray shows Few data and also includes low-confidence findings, marked on the card. They suggest tests and increases, never cuts.
  • When the account does not have enough results to compare, the X-ray suggests how to gather data for the next analysis, for example moving budget to the items that already brought results or testing the best ads.

It reads the pages

The X-ray opens the landing pages that received the most budget. On each one, it checks whether the page opens, whether it loses clicks, whether it works on mobile, whether the form is long, whether there is a button on the first screen and whether the pixel is on it. It also checks what helps conversion: headline, offer, buttons, social proof and trust signals.

It compares the start of the page with the promise of the ad that leads there and warns when the page opens with a different offer. On Google, it judges every search term against what the page offers. A term outside the business that did not convert becomes a negative keyword, and a term that belongs to the business is never made negative.

On the cards, a page appears only as its domain and path, without the parameters after the "?". If the X-ray cannot read a page, it shows why and what to do, such as allowing the Agencify reader in the site's bot protection.

It reads the creatives

For Meta videos, the X-ray compares how many people watch the first 3 seconds with the other videos that pursue the same result. A video that loses many people at the start gets a card with a new version of the ad next to the original.

It also compares the hook, format and offer of the ads inside each ad set, so the effect of the audience is not confused with the effect of the creative. A pattern only counts when it shows up in more than one ad set and still holds without the ad that weighs most on it. Even so, a pattern shows what went together in the period and does not prove the cause.

On Google: impressions, assets and bidding

  • When a Search campaign loses impressions to budget, the X-ray suggests more budget only if it has at least 3 results, costs the same or less than similar campaigns, and lost impressions to budget on 3 days in a row. A budget shared by several campaigns never gets a raise.
  • When it loses impressions to ad rank, the card shows impression share, top of page and click share, and proposes assets and a new ad before any bid change.
  • When sitelinks, callouts or a snippet are missing, the AI writes the missing ones from the ad's page. You review and edit them in the preview, and Agencify creates them and links them to the campaign. You can undo it.
  • When campaigns with the same conversion and a similar target each have few conversions, the card recommends grouping them in a portfolio bid strategy and shows the steps in Google Ads.

Dismissing a card

When you dismiss a card, you choose for how long: a week, a month or forever. You can also stop seeing that type of card for this client. A critical card still shows. In the Dismissed tab, Show again brings the card back.

What it never does

  • In the free analysis, it only reads. Nothing changes in the campaigns.
  • It does not apply a fix without your confirmation, except for the actions you let the AI apply on its own.
  • It does not touch an item someone changed outside Agencify in the last 3 days, such as a change to budget, status, audience, bid or goal.
  • It does not turn back on what someone paused in Ads Manager or Google Ads.
  • It does not take budget from prospecting to give to remarketing.
  • It does not add up results of different kinds.

After you apply

When you apply a suggestion, 7 days later Agencify compares the item before and after and shows what it measured on the card, for example "Cost per lead: from $42 to $31 in 7 days. Improved." If another change to the same campaign falls in those 7 days, the result does not count as measured.

In the Done tab, the scoreboard of the last 90 days counts what improved, stayed the same or got worse, and the estimate shows apart, as an estimate.

Up next

To see how Agencify uses metrics and AI across the app, read How metrics and AI optimization work.

Was this article helpful?