Arcads With Causality Engine: Check the Winner Mark Clones

The short answer: the hook that "won" your Arcads test won on Meta's count, and Meta's count is not the store's. Arcads, the AI video tool whose synthetic actors read your ad scripts, makes the creative; its agent Mark can then "analyze a Meta campaign, clone the winning ad and publish a new Meta ad" (Mark), and Workflows can turn one winning ad into 20 variations in a run (creative testing). This is a partial fit. Arcads reports no sales of its own; it relays what the ad platforms report, and it touches GA4 only through the UTM tags Mark suggests. Check the channel before the clones go out: Causality Engine, a causal attribution tool that reads the store's GA4 export, gives a free first finding at app.causalityengine.ai/start.

Disclosure: I founded Causality Engine and own it, and this page recommends it. Every claim about it links to Causality Engine's own pages.

Three checks before Mark gets the approval click:

1. Count what the winner is worth. For example, a hook with 22 Meta-reported purchases on €300 of spend has a CPA of 300 / 22 = €13.64. Twenty clones at the same budget would spend 20 x €300 = €6,000 on the belief that the 22 were caused by the ad.
2. Read the recap you already get. Before publishing a Meta ad, Mark shows a recap with the copy, the setup it copied from recent campaigns and the UTM tags it suggests; nothing goes live until you approve (connecting accounts).
3. Put a GA4 row beside the test. Paid Social, filtered to the test's campaign, is the last-click view of the same clicks.

What sits where

Whose number picks the winner

Arcads does not count conversions. It makes ads, and once Meta Ads and TikTok Ads are connected under Agent, Connectors, Mark can read "Meta campaigns and ads, TikTok ads, X Ads campaigns" and answer questions such as "How are my Meta campaigns doing?" (connecting accounts). A scheduled routine can send a weekly recap of ad performance (routines). The My ads tab sorts running ads by newest or highest spend and notes that Meta and TikTok show lifetime performance (My ads).

Which conversion metrics Mark reads, and under which attribution setting, the help centre does not say as of 11 November 2026. The source is Meta's reporting, so Meta's rule applies: a website purchase is credited to an ad if it happened within a set number of days after someone viewed or clicked it, shown by default according to the ad set's attribution setting (Meta's attribution system). A hook that earns more of those credited purchases is a stronger hook on Meta. It is not yet proof that the store sold more.

That distinction matters most at the moment Arcads is built for. Its own guide to performance marketing describes connecting Meta Ads so Mark can "read campaign performance, clone the winner and publish new ads". Every step after the word "winner" inherits Meta's count.

The UTM tags in Mark's recap

When Mark publishes a Meta ad, it copies the objective, budget, audience, Page, link and call to action from recent campaigns, writes the copy and suggests UTM tags. Approving puts the ad live and starts spending the daily budget, unless you ask for it paused (connecting accounts). The article names no default values for those tags, so read them on the recap every time.

What GA4 needs from them is short. Paid Social requires a source on Google's list of social sites, such as facebook or instagram, and a medium matching the paid pattern: paid, ppc, cpc or anything starting with "paid" (default channel group). Meta may add source Facebook or Instagram and medium paid on its own, and anything in Ads Manager's URL parameters field overrides a duplicate in the website URL (Meta's URL parameters page). Use one source value per platform, as Google's URL builder page recommends, and a campaign value that names the hook test so its rows can be found later.

TikTok ads from Mark take their landing page from past campaigns, and the article mentions no tags for them. Untagged, they cannot reach Paid Video, the channel whose description names TikTok; a visit whose source is lost lands in Direct (direct traffic).

Worked example: a five-hook test and the store's fortnight

For example, using invented figures and no real account: five hooks from Arcads each get €300 on Meta over 14 days, €1,500 in all. Meta credits them with 9, 14, 22, 11 and 8 purchases, 64 together. Hook C leads with 22.

Over the same fortnight the store books 1,180 orders. Across all campaigns, Meta claims 410, TikTok 296 and Google Ads 590: 410 + 296 + 590 = 1,296, which is 1,296 - 1,180 = 116 more orders than exist. GA4's last click gives Paid Social 133.

Mark now offers to clone hook C into 20 variations. Inside Meta, the ranking is fair: hook C beat hook E by 22 to 8 on equal spend. The open question is a level up: does the Paid Social channel cause anything near the 410 Meta claims, or nearer the 133 GA4 sees? A causal read answers at that level. Causality Engine sets each channel's caused share beside last-click from the same export, with a data-health score from 0 to 100 and a next step, raise, cut or hold, to be tested in one region first (how it works). If Paid Social reads low, keep the hook and shrink the clone run; the creative won, the budget question did not.

Checking for free

By hand, and enough on its own. Keep each hook on the same budget, as the Arcads hook test on this site does, then add a holdout: run the clones in one country and leave a comparable country on the old ads for three weeks. For example, if the clone country's orders rise from 430 to 489 and the other's from 395 to 402, the clones can claim (489 - 430) - (402 - 395) = 52 orders. Ask Mark to create the clones paused, which the article allows, and set the country in Ads Manager before they spend.

When you want all channels judged at once, Causality Engine, which I founded, works from the GA4 Attribution paths export on its start page. The first finding is free and stays in the browser; a full read is €99 once, with a refund inside 30 days (pricing, checked on 11 November 2026).

Limits on both sides

Causality Engine reads channels, never ads or hooks, so it cannot pick between Arcads variations; that stays a platform test. It works from GA4 key events (one-off read), asks for 12 months or more of export (how it works), treats a channel under 1% of credit on both methods as too small to act on, and declines when spend is too small for a signal (homepage). Arcads, in turn, documents no sales report, no GA4 connection and no store connection, so it cannot settle the claim gap either.

Arcads earns from generations: each video spends credits, and Mark's own chat turns are charged by usage (Mark). A winner read off Meta's dashboard, cloned twenty times, is twenty generations paid for on the strength of a number Arcads did not compute and does not check.

See also

• Arcads: AI actors, credits and the disclosure rules
• How to use Arcads: a five-hook test inside one month
• Arcads on the marketplace: Arcads' affiliate program terms
• Causality Engine: its method, prices and refusals
• How to use Causality Engine: export, read, then a one-region test
• Causality Engine on the marketplace: affiliate terms in the owner's words
• Creatify vs Arcads: product-link video or scripted AI actors
• How to export GA4 data for attribution: getting the file a read needs

FAQ

Do Arcads and Causality Engine connect?

They do not. Arcads reads ad platforms through Mark; Causality Engine reads a GA4 export the store owner downloads. The Meta ads Mark publishes appear in that export only as the channel their UTM tags put them in (integrations).

Can Arcads tell me which video sold the most?

Not from the store's side. It can relay Meta's and TikTok's figures per ad through Mark, and the My ads tab shows status, impressions and spend (My ads). Sales per video exist only as the platform attributes them.

Should I accept the UTM tags Mark suggests?

My recommendation: accept them only after reading the medium. If it is paid, ppc or cpc and the source names the platform, GA4 can file the visits as Paid Social; anything else, change on the recap before approving (default channel group).