Which edits reset Meta learning

What counts as a significant edit#

Meta treats targeting, placement, optimisation event, creative and large budget changes as significant: each one restarts the learning clock because the auction must recalibrate to a new setup. Pausing for over seven days has the same effect on restart. Meta lists them in its significant edits guide, and learning phase shows what restarts cost.

Assume any change to who sees the ad, what counts as a win, or what they see is significant. When unsure, batch it with the next planned change rather than testing the boundary weekly.

Safe tweaks that pass quietly#

Name changes, minor bid adjustments within tolerance, small budget nudges and reporting tweaks generally do not restart learning, though they still deserve a changelog entry. The practical rule is proportion: tweaks that leave auction eligibility and optimisation maths untouched pass quietly, while anything that moves predicted action rates restarts the count.

Keep a dated edit log per campaign: what changed, why, and what learning status followed. Patterns across months teach your real thresholds faster than any help article. The hub holds the tracking template.

Creative edits without self sabotage#

Swapping images, video, copy or links inside a live ad resets its learning, so version instead: duplicate the ad, publish the variant beside the control, and let clean A/B reads decide. There is no 20 per cent text rule any more, but keep in-image text short as a performance finding, and check overlay legibility in our overlay guide.

Feed creative range through diversification on a schedule, and retire fatigued ads per fatigue curves rather than editing winners mid flight. Meta creative troubleshooting plus the learning phase guide confirm the reset behaviour.

Budget edits that hold stability#

Scale budgets in steps of 20 to 30 per cent every few days rather than doubling overnight: modest steps usually preserve calibration while jumps restart it. Restructure budget types, for example fixed to campaign budget, only as a dated rebuild with a fresh learning window, never as a casual toggle.

Freeze budgets through the first full cycle after any significant change, and judge on weekly totals. Impatience during ramp is the most expensive edit of all.

An edit discipline that survives scale#

Batch changes into one weekly edit window: targeting, creative and budget moves together, then hands off for seven untouched days. Version major changes as new ad sets or campaigns so controls survive for comparison, and prioritise measurement edits carefully, because with AEM 8-slot configuration the default slot order is unsafe and domain or event changes carry optimisation consequences.

Review Limited flags after each window, not daily. Ask our team for an edit audit when learning never settles.

Rebuilds versus edits#

Where the change list keeps growing, rebuild instead of editing: duplicate the campaign, apply the new targeting, budget and creative together, launch fresh with a full learning window, and keep the original paused for comparison. Rebuilds cost one learning cycle but buy clean reads, while dripped edits cost perpetual learning and muddy history. The structure in first setup plus consolidation gives the rebuild its shape.

Name rebuilds with dates and reasons so audits stay legible, migrate budgets in one move, and freeze the new build for seven untouched days. Reserve rebuilds for quarterly reviews and strategy shifts, not weekly whims. Roll back by unpausing the original when the rebuild loses twice. Learning status confirms the new build has settled.

Frequently asked questions#