The audit and batch workflow prompts in this guide invoke the three SEO Intel skills. Install the Skills Pack before following this guide. Download the Skills Pack →
The audit-fix-verify loop
Every productive store management session follows the same three steps. Audit first: ask Claude to check something specific and return a structured result. Fix second: ask Claude to apply the changes. Verify third: re-run the audit on the changed item and confirm it now passes. Skipping the verify step is the single most common mistake: it takes 30 seconds and catches copy-paste errors, hallucinated rewrites, and edge cases before they go live on your store.
Audit (60 sec) + Fix (2 min) + Verify (30 sec) = 3 to 4 minutes per signal. With 9 signals in the AI Visibility audit and a typical store failing 3 to 4, you can fix everything in a single 20-minute Claude session.
Prompt patterns that work
Audit prompts
Once you have downloaded the Skills Pack and installed the skills, use the prompts below to invoke them.
The key principle: tell Claude what to check, what the pass criteria is, and what format you want the result in.
Run a GMC compliance check on all products in my [collection name] collection. For each product, return a row in a table showing: product title, pass or fail, the specific GMC rule triggered if it fails, and the exact text that caused the failure.
Check my store's robots.txt and tell me which of these AI crawlers are blocked: GPTBot, ClaudeBot, PerplexityBot, Google-Extended, Anthropic-AI, OAI-SearchBot. For each one, show me the exact line in robots.txt that blocks or allows it.
Check whether my Shopify store's homepage has a FAQPage JSON-LD block. If it does, show me the questions and answers. If it does not, tell me what to add and where.
Fix prompts
The key principle: reference the specific audit result, state the exact change you want, and ask Claude to confirm what it changed.
The audit found that [product name] fails the brand-consistency rule. My correct brand name is [brand name]. Update the product title, vendor field, and description to use the correct name consistently. Show me the before and after for each field.
Rewrite the description for [product name] so that the first sentence answers the question "What is this product used for?" in under 20 words. Keep the rest of the description but do not change the price or any specification data.
Add a FAQPage JSON-LD block to my homepage with 5 questions and answers based on my most common customer queries. The questions should be about [topic 1], [topic 2], [topic 3], [topic 4], and [topic 5]. Return the complete JSON-LD I need to paste into my theme.
Verify prompts
Re-run the brand-consistency check on [product name] using the updated version. Confirm whether it now passes and show me the specific field values that were checked.
Check the FAQPage JSON-LD I just added. Validate that it is syntactically correct JSON-LD, that it uses the schema.org FAQPage type, and that each question-answer pair is in the correct format. Return any errors found.
Batch workflow prompts
With your store connected via MCP, Claude can paginate through collections automatically. These prompts are the most time-saving of all. Once you have downloaded the Skills Pack and installed the skills, use the prompts below.
Go through every product in my [collection] collection. For each one, check whether the description contains a sentence that answers "What is this product?" in the first 60 words. Return a list of the ones that fail this check, sorted by the ones most likely to affect AI visibility.
Check all products in my store for brand-name consistency. My store name is [name], my brand is [brand]. Flag any product where the vendor field, title, or description uses a different spelling or capitalisation. Return a table with the product title and the inconsistent text found.
Prompt mistakes to avoid
Vague scope: "Check my store" returns a surface-level overview. "Check the robots.txt on my store for AI crawler blocks and list every User-agent rule" returns a specific, actionable result.
No output format specified: Without a format instruction, Claude returns flowing prose. Adding "return a table" or "return a numbered list" makes results directly actionable.
Missing context: If you have not pasted your store boilerplate at the start of the session, Claude does not know your brand name, your compliance history, or your pricing rules. Always start with the boilerplate. See Guide 4.
Skipping verify: Claude occasionally makes small errors in rewrites, particularly with pricing and specification data. Always verify after fixing, especially before pushing changes to your live store.
Combining the three skills in one session
Once you have downloaded the Skills Pack and installed all three skills, the most productive session structure is: (1) Paste your store boilerplate. (2) Say "run an AI readiness audit" to invoke the GEO / AI Visibility Skill and get your AI Readiness Score. (3) Say "run a GMC compliance check" to invoke the GMC Compliance Skill and get the pass/fail table. (4) Say "run an AEO check" to invoke the AEO Skill for content rewrite suggestions. (5) Prompt Claude to prioritise all fixes across the three results by impact. (6) Execute fixes in priority order with the fix prompt patterns above. (7) Verify each fix before moving to the next. This sequence covers your entire AI and GMC compliance posture in one sitting.
Sources and references
- Anthropic: Prompt engineering overview: Anthropic's official guide to writing effective Claude prompts
- Shopify Admin GraphQL API: The API Claude queries when you ask for store data
- Guide 2: Use the Skills: Install and run the three skills before building your prompt workflow
- Guide 4: Build Your Store Boilerplate: The context file that makes every prompt more effective
- AI Visibility Playbook series: Signal-by-signal detail on what the AEO skill audits
- Free AI Visibility audit: Run this first to establish your baseline ARS score
- Free GMC audit: Companion baseline check for Google Merchant Center compliance
Frequently asked questions
What makes a good prompt for store management tasks?
Specificity and context. A good prompt states the task (audit, fix, verify), the scope (one product, a collection, the whole store), the desired output format (table, list, rewritten text), and any constraints (do not change prices, match the brand voice in the boilerplate). Vague prompts return vague results.
What is the audit-fix-verify loop?
It is the three-step workflow that produces reliable results: (1) Audit: run a skill or ask Claude to check a specific signal. (2) Fix: prompt Claude to apply the fix to the relevant product or page. (3) Verify: re-run the audit on the changed item and confirm the signal now passes. Never skip the verify step. It takes 30 seconds and catches errors before they go live.
How do I run a batch product audit without doing one product at a time?
With your Shopify store connected via MCP, prompt Claude: 'List all products in [collection name] and check each one against the GMC misrepresentation rules. Return a table showing product title, pass or fail, and the specific rule triggered for any failures.' Claude will paginate through the collection automatically.
Can I save prompt sequences for reuse?
Yes. Save your most-used prompt sequences in your store boilerplate (covered in Guide 4) or as a text file you paste at the start of each session. You can also create custom skills that embed your standard prompts as the default behaviour when you invoke them.