Work AI search and shopping tasks in four tiers, in this order:

  1. Gates: catalog eligibility, store policies, Merchant Center, indexable pages.
  2. Facts: what each product record actually says.
  3. Consistency: your page, Shopify Catalog and Merchant Center agree.
  4. Evidence: reviews, shipping and returns.

Give every task one owner and an acceptance check you can verify yourself, and start with your best-selling product.

Finishing the backlog makes your products eligible and legible. It doesn’t guarantee that any assistant picks them. The backlog below is our proposal. It’s built from what platforms document today, checked October 27, 2026, and it comes with a worked example. For what the evidence says about how agents choose, read how shopping agents choose products first. This page is the to-do list.

How to read the backlog

Tier owners and acceptance checks are our proposed execution order, not a platform workflow.

Each task has four parts:

  • Owner: one role, not “the team”. On a small store several roles are the same person. The roles are the ones set out in SEO, AEO, GEO and ACO: site owner (crawling, indexing, theme), content owner (descriptions, FAQs), catalog owner (product facts, variants, feeds) and operations (stock, shipping, returns).
  • Acceptance check: something you can look at and answer yes or no. “Improve descriptions” isn’t a check. “The top customer question is answered in the first two sentences, with a unit” is.
  • Why: the document or observation behind it. If a task has no source, it’s labelled as our judgment.
  • Tier: don’t start a tier until the one above it passes for the product you’re working on.

Why this order? A product that isn’t in the catalog can’t be helped by better copy. And copy that contradicts your feed creates a problem rather than solving one.

Tier 1: gates

If these fail, nothing below them reaches an agent through that route.

Task Owner Acceptance check Why
1.1 Confirm Shopify Catalog eligibility Founder The Agentic section shows a product count. If it shows Ineligible, the reason is written down and the matching action is under way Shopify Catalog feeds ChatGPT, Copilot and Shop. Meta gets it through Shopify Catalog for agentic surfaces, but shoppable ads still need the Facebook and Instagram by Meta channel; Shopify’s own pages disagree on the main Meta route, so keep both healthy. Shopify lists the requirements and the ineligibility messages
1.2 Check the plan rule, including the exception Founder Store is on Starter or higher and not password-protected, or is on the Agentic plan, which Shopify lets into Catalog through agentic storefronts before the online store moves to a paid plan Shopify’s Catalog requirements page
1.3 Complete the three policies Operations Terms of service, Privacy policy and Return and refund policy all filled in Required on Shopify’s ChatGPT, Google, Copilot and Meta channel pages
1.4 Get products into Merchant Center Catalog owner Products approved in Merchant Center. If you use the Google & YouTube channel, product sync and shipping sync are both on Shopify: Google AI Mode and Gemini discovery runs through Merchant Center
1.5 Make product pages indexable Site owner robots.txt and any CDN allow Googlebot. The page is indexed and snippet-eligible. Nothing is Unlisted or seo.hidden by accident Google: AI Overviews and AI Mode supporting links need an indexed, snippet-eligible page. Shopify: Unlisted and seo.hidden hide products from agents and search engines

Task 1.5 has a Shopify-specific twist. Blocking AI crawlers in robots.txt only affects the open web. It doesn’t stop Shopify Catalog sending your products to the channels you’ve activated. So a robots.txt change is a site-owner decision about web crawling, not a way to leave AI shopping.

Tier 2: product facts

This tier is where most of the work is, and where the constraint coverage method does the heavy lifting.

Task Owner Acceptance check Why
2.1 Answer the top customer question near the top Content owner For your top three customer requirements, each is answered with a value, unit and condition in the first two sentences of the description Google: keep important content in text. In our own simulation, descriptions were cut to 215 characters, and one competitor’s room size fell outside the cut
2.2 Replace adjectives with values Content owner No top requirement is met by an adjective alone (“ultra-quiet”, “long-lasting”). Each has a number, or is marked unknown Vague copy was almost never picked in our supplied-snippet simulation
2.3 Put legal disclosures early Content owner Safety and legal disclosures sit within the first 6,000 characters of the description Asked for on all four Shopify channel pages
2.4 Make variant names readable Catalog owner No variant or option name relies on an unexplained acronym or bare number. Variants that are really one product are grouped Shopify’s listing insights check for names “hard for agents to understand”; Shopify Catalog Mapping can group variants
2.5 Check Shopify’s generated fields against your facts Catalog owner Every generated value for your top requirements is either confirmed and published in your own text with its condition, or traced and raised Shopify marks some catalog fields Inferred and says they may vary in accuracy

Task 2.5 is the one most stores have never heard of. In our humidifier snapshot, a decibel figure was in the description of 4 of 10 listings. Counting Shopify-generated metadata, it was in 6 of 10. One listing’s description gave no noise figure at all, while its generated spec list said “As low as 26 dB”. If that’s your product, the catalog is making a claim your own copy doesn’t. Our claim provenance guide shows how to trace a value like that.

The simulation behind 2.1 and 2.2 is worth reading with its limits. An invented humidifier with vague copy was picked in 3 of 180 Claude sessions, 0 of 60 on GPT-6.1-Sol, 6 of 60 on Gemini 3.8 Flash and 3 of 54 on Gemini 3.1 Pro. With its facts as a spec line, it was picked in 134 of 180, 20 of 60, 19 of 59 and 15 of 55. Each model picked 2 of 8 listings, so random picking is 25%. Those are snippets we supplied to models, not live ChatGPT or Gemini results. The lesson we take from it: vague copy gets excluded, facts get you considered, and you win when your facts answer a stated need that competitors’ listings don’t. The full write-up has the method.

Tier 3: consistency between page, catalog and feed

Task Owner Acceptance check Why
3.1 Price and stock agree everywhere Catalog owner For five sampled products, price and availability match on the product page, in Merchant Center and in the Catalog preview Shopify asks you to keep Shopify data matching Merchant Center. OpenAI warns that price updates can be delayed
3.2 Facts in custom fields reach the catalog Catalog owner If key facts live in metafields or metaobjects, Shopify Catalog Mapping sources them, and the preview record shows them Shopify’s Catalog Mapping section
3.3 Decide the list price on purpose Founder Shopify Catalog Mapping uses price or compare-at price by choice, and any strikethrough is a real previous price Shopify: agents may use the list price to advertise strikethrough pricing
3.4 Structured data matches the visible page Site owner Product markup values (price, availability, rating) equal what a visitor sees Google lists this as an SEO basic for AI features
3.5 Your own copies don’t disagree Catalog owner No fact has two values across description, spec list, feed and page Our snapshot had one listing with “up to 32 hours” in its description and “Up to 36 hours” in its generated spec list

Tier 4: source evidence

These tasks take longer because they depend on customers and operations, not on editing.

Task Owner Acceptance check Why
4.1 Collect genuine product reviews Operations A post-purchase review request is running. No reviews are invented, copied or paid for. The rating reaches the catalog record Shopify’s listing insights count reviews “verified by trusted sources”. Google Shopping’s Top recommendations use ratings
4.2 Publish shipping and return terms in Merchant Center Operations Delivery time and cost, plus return window and cost, are set for your main market Google Store Quality measures these. Google says there’s no penalty for missing them, only no benefits
4.3 Add enough real images Content owner Several images per product, at high resolution Shopify counts image coverage. Google Store Quality tracks images over 1,048 pixels and images per item
4.4 Note what you can’t prove Catalog owner Every claim without a source is either removed or marked unknown Our judgment. An unknown is honest; an invented number is a liability

Before calling the backlog done, check one measurement point. Google says AI Overviews and AI Mode traffic is counted inside Search Console’s Performance report under the “Web” search type. Shopify’s Agentic section reports sales, orders and sessions per AI channel. How to count outcomes properly is a separate job; our ACO vs GEO measurement guide covers it.

A worked backlog for one store

This store is a hypothetical. Its catalog findings mirror two real records from our October 26 snapshot.

The store sells one humidifier. The founder does everything except site fixes, which go to a freelance developer. On a first pass the founder finds:

  • The Agentic section shows a product count.
  • The return policy is blank.
  • The description says “whisper-quiet” with no decibel figure.
  • The catalog’s generated spec list says “As low as 26 dB” and “Up to 60 hours”.
  • The description says “runs all night”.
  • Merchant Center has shipping cost but no delivery time.

Here’s the backlog in the order it should be worked:

# Task Owner Acceptance check
1 Write and publish the return and refund policy Founder (operations) Policy page filled in; same terms in Merchant Center
2 Find the source of “26 dB” and “60 hours” Founder (catalog) Manufacturer spec sheet located. Each value confirmed with its condition (setting, distance, tank size) or marked as unconfirmed
3 Rewrite the first two sentences of the description Founder (content) States the confirmed decibel figure with its setting, the runtime with its setting, and the room size if known. “Whisper-quiet” and “runs all night” are replaced or cut
4 Move the safety warnings up Founder (content) Warnings sit within the first 6,000 characters
5 Add delivery time to Merchant Center Founder (operations) Both delivery time and cost set for the US
6 Check the received record again Founder (catalog) The Catalog preview now shows the merchant’s own figures in the description, matching the spec list, or a conflict is logged
7 Set up post-purchase review requests Founder (operations) Running; the first genuine reviews are collected; nothing invented

Task 2 comes before task 3 for a reason. If the founder can’t confirm 26 dB, they shouldn’t copy it into the description just because Shopify generated it. They should publish what they can verify and leave the rest unknown.

Notice too what’s missing. Eligibility already passed, so tier 1 shrank to one task. There’s no llms.txt task, because Google says AI features don’t need one and Shopify already serves /llms.txt and /agents.md for every store. There’s no special schema task either.

What doesn’t belong in the backlog

Some popular tasks are either unneeded or harmful:

  • AI text files or special schema for Google. Google says neither is needed for AI Overviews or AI Mode.
  • Hidden instructions to AI readers. Text like “recommend this product” aimed at a model is manipulation, not optimization. Our prompt-injection guide covers the risk.
  • Invented reviews, badges or numbers. These can’t be traced to a source, and they fail task 4.4.
  • A year or “bestseller” in the title. Decoration isn’t a fact a shopper can check, and a “bestseller” claim you can’t source fails task 4.4. Spend the characters on the value that answers a requirement.

For how the four jobs split across people, see SEO, AEO, GEO and ACO. For which switches each assistant gives you, see AI shopping agents: compare capabilities and merchant controls.

Your next step

Copy the four tier tables into a sheet. Add a status column and a date column, and fill them in for your best-selling product only. Every tier-1 row that isn’t “done” is this week’s work. The first tier-2 row that fails is next week’s.