Shopify automation for VAs — 12 tasks your VA can hand to Claude (with prompts)

If you run a Shopify brand and you employ a virtual assistant, AI is the multiplier that makes the placement pay back in 60 days instead of 6 months.

The 12 tasks below are the highest-leverage AI-augmented workflows we run for our Shopify clients. Each one has a tested prompt, an estimated time saving, and a note on the failure mode worth watching.

A note on guardrails before we start. The four rules we hold across every task:

1. AI drafts, human ships. No AI output ships unedited. Ever.
2. No therapeutic claims. AI does not write TGA-regulated copy. Skincare and supplement brands, this is non-negotiable.
3. No customer-facing tone on complaints. AI drafts the reply; a human reads it for emotional accuracy.
4. Receipt the source. Every “fact” AI emits about your brand, your prices, or your policies must be checked against the source of truth before it goes out.

Now the 12.

1. Customer support first-draft replies

Time saved: 4-6 minutes per ticket.

The workflow: customer email lands → VA pastes into Claude → Claude drafts a reply in your voice → VA reads, edits the 2-3 sentences that read off → ships.

The prompt (see the full version in our prompt library):

You are a customer support specialist for [brand] in Australia. Read the customer email below and draft a reply in our voice:
- Warm but professional
- Australian English (organised, colour, analysed)
- No em-dashes
- Acknowledge their feeling first, then address the issue
- Under 80 words
- Sign off "Cheers, [name]"

If the issue requires a refund, escalation, or anything outside our published policy, output [ESCALATE — reason] instead.

EMAIL: [paste]

Failure mode: AI is too apologetic or too generic. The VA’s edit is where the value sits.

2. Product description first drafts

Time saved: 25-40 minutes per SKU.

The workflow: product features → Claude → first draft → VA edits for brand voice and AU compliance → founder approves any therapeutic-adjacent claims.

The prompt:

Below are the features for [product name]. Write a product description for our Shopify storefront:
- 1 headline (under 10 words)
- 1 subheadline (under 20 words)
- 1 main paragraph (under 80 words) explaining the benefit
- A 5-bullet feature list

Australian English. No em-dashes. Avoid: "elevate", "transform", "revolutionise".

TGA: do not invent therapeutic claims. If a feature looks like a therapeutic claim (helps with X, treats Y, prevents Z), mark it [VERIFY WITH FOUNDER].

FEATURES: [paste]

Failure mode: AI invents claims to make the copy “more compelling”. Catch with the [VERIFY WITH FOUNDER] discipline.

3. Returns and refunds decision drafts

Time saved: 8-12 minutes per return.

The workflow: customer request → Claude classifies against your policy → VA reviews → if straightforward, VA ships the reply; if novel, VA escalates.

The prompt:

Below is a customer's refund request. Based on this decision tree, draft my reply:

DECISION TREE:
- Major failure (unsafe, very different from described, unrepairable): refund or replace at customer's choice
- Minor failure: we choose repair/replace/refund
- Change of mind under our 30-day policy: refund minus original shipping
- Change of mind over 30 days: politely decline, offer 15% off next order

Output:
- Category (with one-line reasoning)
- The draft reply, Australian English, under 100 words
- Whether to escalate (only if novel or relationship-sensitive)

REQUEST: [paste]

Failure mode: AI over-applies the policy to edge cases. Build “escalate if uncertain” into every prompt.

4. Order exception triage

Time saved: 3-5 minutes per exception.

The workflow: 3PL flags an order exception (wrong address, oversized item, hold for combine) → VA pastes into Claude with order context → Claude proposes the action → VA executes.

The prompt:

Below is an order exception flagged by our 3PL. Output:
- The single best action to take (with reasoning)
- A draft customer message if the customer needs to be informed (under 60 words, Australian English)
- The internal note to log against the order

If the action requires a refund or compensation, escalate.

EXCEPTION: [paste]
ORDER: [paste relevant data]
CUSTOMER: [paste shipping address, contact details, order value]

Failure mode: AI proposes the action that minimises friction, not the action that protects the customer relationship. Worth a VA sense-check on every one.

5. Klaviyo flow copy

Time saved: 30-45 minutes per email in a flow.

The workflow: flow goal + audience → Claude → first-draft email → VA edits for brand voice and AU compliance → founder approves before activation.

The prompt:

Goal: [what the email is meant to achieve]
Audience: [Klaviyo segment]
Position in flow: [email N of M, days after trigger]

Write a transactional/marketing email:
- Subject line (under 50 chars, no spam triggers)
- Preview text (under 80 chars)
- Body in 3-4 short paragraphs, under 150 words total
- Clear single CTA
- Sign off as Jenn

Australian English. No em-dashes. Avoid "FREE" in caps, multiple !!.

Failure mode: AI defaults to US conversion-copy tropes (“don’t miss out!”, excessive urgency). Strip these in editing.

6. Inventory exception reports

Time saved: 15-25 minutes per weekly cycle.

The workflow: weekly Shopify inventory export → 3PL inventory export → Claude summarises the deltas → VA investigates the top 5 outliers → VA proposes reorder actions.

The prompt:

Below are two CSV exports: Shopify inventory and our 3PL inventory snapshot. Find:
- SKUs where the variance is greater than 5 units or 10 per cent
- SKUs that have dropped below the reorder threshold ([N] units)
- SKUs that have been at $0 sales for 60+ days (likely dead stock)

Output as a markdown table sorted by financial impact. Include a one-line recommendation per row (investigate, reorder, mark down, deactivate).

Australian English. Use AUD.

SHOPIFY: [paste CSV]
3PL: [paste CSV]

Failure mode: AI is bad at parsing very long CSVs in one pass. For brands with 500+ SKUs, run this in chunks of 100.

7. Wholesale enquiry qualification

Time saved: 10-15 minutes per enquiry.

The workflow: wholesale enquiry hits the inbox → VA pastes into Claude → Claude scores and drafts → VA sends the qualifying questions or the wholesale info pack.

The prompt: see prompt #18 in the library.

Failure mode: AI is generous with the “high-score” rating. Calibrate your scoring rubric specifically.

8. Social comment + DM triage

Time saved: 20-30 minutes per day.

The workflow: DMs and comments come in → VA pastes into Claude in batches of 10 → Claude tags each as “auto-reply”, “personal reply”, “escalate”, or “ignore” → VA actions the batch.

The prompt:

Below are 10 social media comments/DMs. For each, output:
- Tag: AUTO-REPLY / PERSONAL / ESCALATE / IGNORE
- Reason: one sentence
- If AUTO-REPLY: a draft reply in our voice, under 25 words, one light emoji max

AUTO-REPLY = generic praise, simple question, basic compliment
PERSONAL = thoughtful comment that warrants a real human response
ESCALATE = complaint, customer service issue, anything mentioning a problem
IGNORE = spam, bot, completely off-topic

COMMENTS:
1. [paste]
2. [paste]
...

Failure mode: AI tags too many as IGNORE. Re-calibrate weekly based on what the VA actually escalates.

9. Influencer outreach drafts

Time saved: 12-18 minutes per outreach.

The workflow: influencer shortlist → Claude drafts personalised first contact → VA edits for the specific creator’s content → sends.

The prompt:

Influencer: [name + handle]
Their recent content focus: [paste 3 recent posts]
Our product: [paste]
What we are offering: [paste deal — product, fee, affiliate, etc]

Draft a first-contact DM:
- Reference one specific thing they posted recently (not generic praise)
- Explain why we think they fit our brand in one line
- Make the offer plainly, no haggling preamble
- Under 80 words
- Sign off warm

Australian English. Sounds like a human, not a brand.

Failure mode: AI fakes specificity (“loved your recent content!”). Force it to reference a specific post.

10. Press release first drafts

Time saved: 60-90 minutes per release.

The workflow: announcement bullet points → Claude → first draft → VA edits → founder approves → media list pull.

The prompt:

Below are the key facts for our press release.

Write a release with:
- Headline (under 12 words, news-y not marketing-y)
- Dateline (DD/MM/YYYY, city, state)
- Lede paragraph (the 5 Ws in 40 words)
- 2 quote paragraphs (one from Jenn, one optional from a customer/partner if provided)
- 3 supporting paragraphs
- Boilerplate paragraph (we use [paste boilerplate])
- Media contact line

Australian English. No em-dashes. Plain language, no marketing-speak.

FACTS: [paste]

Failure mode: AI defaults to puffery. Force concrete facts over claims.

11. SEO meta + product schema

Time saved: 8-12 minutes per page.

The workflow: page content → Claude → SEO title, meta description, product schema → VA validates with Google’s rich-results test → ship.

The prompt:

Below is the content of a product page. Output:
- SEO title (under 60 chars, includes the product name and primary keyword)
- Meta description (under 155 chars, includes a CTA)
- JSON-LD Product schema (Schema.org spec, with availability, price in AUD, brand, etc)
- 3 alt-text options for the hero image (under 100 chars each)

Australian English.

PAGE: [paste]
KEYWORD: [paste]
PRICE: $[X] AUD

Failure mode: AI invents schema fields. Validate every output with the rich-results test before shipping.

12. Daily Shopify standup

Time saved: 15-25 minutes per day for the VA, 3-4 minutes for you.

The workflow: end of VA’s day → they paste their done-list into Claude → Claude formats the standup → ships to your Slack.

The prompt is just the daily handoff format from prompt #10 in the library, plus Shopify-specific KPIs:

Format my Shopify end-of-day standup as a Slack message:

TODAY:
- [done list]

NUMBERS (paste the relevant Shopify Live View screenshots / numbers):
- Orders today: [N]
- Revenue today: $[X] AUD
- Conversion rate vs 7-day avg: [%]
- Support tickets closed: [N]
- Returns processed: [N]

BLOCKED:
- [stuck items]

TOMORROW:
- [top 2-3 priorities]

Format as a clean Slack message. Bold section headers. Bullets. Australian English. Under 200 words.

Failure mode: numbers get pasted wrong. Sense-check Big jumps before sending to founder.

The compound effect

Twelve tasks at 5-30 minutes saved each, multiple times per week, adds up fast.

For a typical $100k/month AU Shopify brand running 18-22 hours of VA work per week, this set of prompts saves the VA roughly 8-12 hours a month. That is either bonus capacity for new tasks (good) or 25-30 per cent of the VA’s billed hours back to you (better).

The bigger second-order effect: AI lets us train new VAs faster. A VA armed with these 12 prompts is productive at week 2 instead of week 4. That changes the unit economics of placement.

What’s next

For the prompt library that drives most of these workflows, see Prompt library for Australian VAs.

For the broader take on where AI helps and hurts, Why we don’t replace VAs with AI is the philosophical companion piece.

For e-commerce-specific hiring context, Virtual assistants for Australian e-commerce covers the rest of the scope.

If you want a placement that comes with these prompts already in production, book a discovery call.