Auto-pick

Not sure which model belongs at a placement? Turn on auto-pick. The app tries the models, counts the clicks, and favors what works — on its own.

Auto-pick needs the Growth plan or higher. It is the Auto-pick tab on the Recommendations page in the PersonalizerAI dashboard. On a lower plan the tab shows which plan includes it.

How it works

Auto-pick works per placement. Once a day, at 01:30 UTC, the app updates its estimate of each model's click rate. It then serves the strongest model most of the time, and keeps testing the others a little. Old evidence fades slowly, so the app adapts when your store changes.

Auto-pick chooses only for blocks whose Recommendation model is set to Automatic (best for this spot). The block's Placement must be the placement where auto-pick is on.

Four precedence facts:

• A block where you chose a model always shows that model. Auto-pick does not change it.
• A block set to Automatic shows the suggested model for its placement while auto-pick is off. When auto-pick is on, the block shows the model that auto-pick chooses.
• A running A/B test that pins a model beats auto-pick. A Model comparison test (or a custom test that sets a model) covers every recommendation spot on your store, not just one. While it runs, shoppers in its test variants see the model the test assigns them, everywhere. Shoppers in its control group, and shoppers the test leaves out, still get auto-pick. Auto-pick does not learn from the test variants' clicks, because the test chose those models, not auto-pick. The Auto-pick tab names the running test so you know. See A/B testing.
• A Frequently Bought Together bundle block always shows its own model. Auto-pick does not change it.
• On the day you turn it on, the app has no click history for that placement yet, so it starts by giving each candidate model an even chance and learns from there. It does not sit idle waiting for the next overnight run.

Turn auto-pick on

Auto-pick needs two things: a block set to Automatic, and the switch for its placement.

1. In the Shopify theme editor, select the Recommendations block, the Cart recommendations block, or the PersonalizerAI app embed (for the cart drawer).
2. Set Recommendation model to Automatic (best for this spot). A new block starts on this setting.
3. Set the block's Placement to the spot you want auto-pick to run. Auto-pick does not run on a block with Automatic — match the page. The cart drawer strip always uses Cart drawer.
4. Save the theme.
5. Open the PersonalizerAI dashboard.
6. Go to Recommendations and open the Auto-pick tab.
7. Select the placement in the Placement dropdown.
8. Turn on the switch on the placement's card.

The card names the recommendation types auto-pick chooses between for that placement — the ones that fit it and that you have turned on in the Models tab. Under the name, one line says where things stand:

• Off — blocks set to Automatic show the suggested recommendation type for the placement. The line names that type.
• On, collecting data — results update once a day at 01:30 UTC, so the first winner shows the day after shoppers start seeing recommendations there.
• Currently winning: … — the recommendation type getting the most traffic in the last 7 days, with its click rate.
• This spot isn't set up for your store yet — you should not normally see this. Every store starts with a row for all nine placements in the dropdown. A store installed before a placement existed gets it the next time the app reconnects to Shopify; if the message stays, contact support.

The checkout, post-purchase, thank-you, and order-status surfaces are not in the list: they are drawn by Shopify's own extensions and send no click data, so there would be nothing for auto-pick to learn from.

The Placement dropdown lists the same nine placements as the block's Placement setting in the theme editor — see Widget catalog — and in the same order.

The Auto-pick tab: placement dropdown, the switch, and what is winning.

Read the details

Open Details (impressions, daily log) on the placement's card to see what the app picked. It appears once the placement has data.

Last 7 days

One row per recommendation type at the selected placement:

A large share for one type means the app has found a clear leader. Even shares mean the app is still exploring.

Daily log (last 14 days)

One row per day and recommendation type: Day, Recommendation type, Impressions, Clicks, and Click rate. Both tables have a CSV button.

The daily log: one row per day and recommendation type.

How long it takes to settle

Be patient with auto-pick. The estimates update once a day, so nothing changes in the first hours. A placement with thousands of daily views separates a winner within days. A placement with little traffic can need weeks. The shares never reach 100% — the app always keeps testing a little, on purpose. When you turn the switch on, the first data appears after the next nightly update.

What happens on downgrade

On a plan below Growth, the app stops auto-picking, and the Auto-pick tab shows which plan includes it instead of the switch. Each block set to Automatic then shows the suggested model for its placement. Each other block shows its own model, as before.