★ FROM OUR SPONSOR

Turn Your Amazon Best Sellers Into 4X Bigger Shopify Orders

Your Amazon best sellers have already proven themselves, yet most sellers only sell them on Amazon. In this Seller Session, Seth Stevens (Seller Growth Network) and Ishan Soni show you how to copy your winning products onto Shopify and put them in front of buyers who are already searching.

The payoff is 4X–6X bigger orders, much better margins because Amazon takes no cut, and you own the customer. You’ll learn how to reuse the Amazon PPC data you’ve already paid for, launch traffic from $20/day per product, run campaigns without constantly making new creatives, and optimize everything in under 15 minutes a month. You don’t need new products or a new business model. You just turn the assets you already have into a profitable second channel. Most sellers treat Shopify like Amazon. Don’t be one of them.

Creating Instagram Ads with AI, From Competitor Research to On-Brand Creative

The opportunity goes beyond making attractive images. AI can help you investigate how other brands sell, spot customer needs your product can credibly address, and turn those findings into a manageable set of original ads to test.

The workflow runs in five stages: brand brief → competitor evidence → creative hypotheses → image production → paid testing.

This guide covers research with ChatGPT's Astra model and image creation inside ChatGPT. Two other routes, Higgsfield and pre-launch checks, are covered in the full SOP linked at the end.

What You'll Need

  • ChatGPT with GPT-6 Astra in your model selector. Browser access helps with live research. Image generation is a separate capability used for production. Access and limits depend on your account and workspace.

  • The public Meta Ad Library. The manual research route needs no advertising-account connection.

  • Your brand and product assets: guidelines, logo files, original product photos, approved claims, your current offer, landing page and genuine customer feedback.

  • Optional: a Higgsfield account. Its connected ChatGPT workflow requires a paid subscription and uses generation credits.

  • A working folder and evidence table to keep references, briefs, approved assets and results together.

Start with one product, one market and one campaign objective. The sample sizes below are starting points, not statistically representative market research.

Step 1: Build a Usable Brand and Product Brief

Gather the source material. Upload brand guidelines alongside product photos and offer details. If you have no guidelines, provide examples from your website, packaging and social posts. Keep three kinds of reference separate:

  • Your brand assets define visual identity and voice.

  • Your product facts and approved evidence define what the ad may claim.

  • Competitor ads show communication techniques. They are not permission to borrow anyone's identity or claims.

Ask Astra for a one-page brief:

Create a practical creative brief for Instagram ads using the attached materials.

Brand: [name]

Product: [name and URL]

Market and language: [details]

Campaign goal: [purchases / qualified leads / awareness]

Audience and buying situation: [details]

Current offer and landing page: [details]

Include:

1. Positioning and primary customer need.

2. Approved benefits, with the supporting source for each.

3. Exact brand colors and fonts where supplied; mark inferred details as provisional.

4. Voice, example phrasing and language to avoid.

5. Composition, photography and logo rules.

6. Product details that must remain accurate: shape, color, label, size and function.

7. Approved offer terms, customer quotes and any mandatory wording.

8. Missing information that blocks accurate production.

Use the formal guidelines as authoritative. Do not invent claims, reviews,

certifications, discounts or product features. Keep competitor references separate.

Do not generate images yet.

Approve it before moving on. Check the brief against your originals. A font inferred from a screenshot is not a confirmed font, and a customer comment is not evidence for a clinical claim. Save the approved version as Brand and Product Brief v1.

Step 2: Research Competitor Ads

Define your search. Start with five direct competitors and two adjacent brands addressing a similar buying situation. Note the target country, language and product category. Aim for 20–30 distinct creatives, mixing older active ads with recent ones. Studying only older ads can miss newer ideas and overrepresent evergreen campaigns.

Work through the Meta Ad Library:

  1. Open the Meta Ad Library, select your target country and All ads.

  2. Search one advertiser by name and verify it's the correct advertiser page.

  3. Narrow to Instagram and your language. For this guide, start with images and inspect video separately if useful.

  4. Open individual ads to record the library ID, link and visible information.

  5. Repeat for the other advertisers. Keyword searches can surface extra examples, but don't treat them as exhaustive.

Record the filters you actually used, since controls vary by market and interface.

Let Astra assist through the browser. Select Astra in a supported ChatGPT interface and ask it to use the browser. Browser interaction and screenshot-based inspection are documented by OpenAI, but that doesn't guarantee access to every Meta page, and ordinary web search alone is not a complete library audit.

Use the browser to research the public Meta Ad Library at

https://www.facebook.com/ads/library/.

Product category: [category]

Target country and language: [details]

Advertisers: [names and page links if known]

Goal: identify original creative directions worth testing for [our product].

First inspect three ads to confirm that you can access their actual creative

and ad details. Report any missing fields or access restrictions.

Then collect up to 25 distinct image creatives across these advertisers,

including older active and recently started examples. Apply Instagram and

language filters where available. Record the filters and collection date.

For each, capture: advertiser, library ID, source link, visible start date,

status, platforms, headline/hook, visual description, selling angle, offer,

proof used, CTA and landing page if accessible. Record reach only if displayed,

with its exact label, geographic scope and reporting period if provided.

Distinguish facts visible in the ad from your interpretation. Group near-identical

versions into creative families. Do not invent spend, CTR, conversions or ROAS.

Do not label an ad a proven winner. Do not infer a video from its thumbnail.

Stop at access restrictions; tell me what screenshots or details to supply.

Return an evidence table and a coverage note. Do not generate or publish ads.
If Meta blocks access, requires human verification or fails to load creative, switch to the manual fallback. Screenshot the full ad plus its details, copy the library link and ID into your evidence table, label the files (for example BrandA_AdID_creative and BrandA_AdID_details), then upload a small batch and ask Astra to analyze only what you supplied. For video, provide viewable footage or key frames plus a transcript. No special "Meta Ads Library MCP" is required, and a Higgsfield connection does not grant access to competitors' private ad results.

Keep one row per distinct creative in your evidence table, grouped into these field types:

Field group

Record

Traceability

Advertiser, library ID, link, screenshot filename, collection date

Delivery context

Selected country, visible start date, status, platforms

Message

Hook, benefit, objection addressed, offer, proof, CTA

Design

Format, hierarchy, product visibility, setting, text density

Transparency data

Exact metric label, value, scope and period if available; otherwise "not available"

Interpretation

Creative family, possible mechanism, relevance to your product, uncertainties

Step 3: Turn Examples into Testable Ideas

Ask what each creative helps the customer understand or believe. A product diagram might explain an unfamiliar feature, a real review might answer a purchase objection, or a lifestyle image might make a use case instantly clear.

Analyze the evidence table and attached creatives against our approved brief.

Identify five recurring creative patterns. For each, provide:

- The customer question or objection it addresses.

- Supporting ad IDs and links from our sample.

- The observed message and visual technique.

- A hypothesis about why the technique might help; not a performance claim.

- What is overused or missing in this sample.

- An original application for our brand using only our approved facts.

Recommend three conceptually different ads for our next test.

For each give the audience situation, angle, exact proposed headline,

primary text, visual layout, approved proof, CTA and landing-page match.

Rank by relevance, clarity, credibility and distinctiveness. Explain the ranking;

do not manufacture a predicted conversion score. Do not copy competitors' assets,

testimonials, slogans or distinctive branding.

Three useful starting formats:

Format

What to create

Evidence needed

Feature-to-benefit diagram

A clear product image with two or three useful callouts

Verified product features and benefits

Customer proof card

Product image, short authentic quotation and restrained design

Exact approved quote and attribution; verified rating if used

Use-case or offer card

One recognizable situation or one clear reason to buy now

Accurate use case or real offer and terms

For example, a work bag concept could show an approved laptop size fitting into its compartment, which demonstrates a verified feature. Don't add a water bottle pocket, a waterproof claim or a capacity your actual bag doesn't have.

Select three concepts before production. A new background color alone is not a new strategic concept.

Step 4: Create the Images in ChatGPT

Use this route when you want to research, write and iterate in one conversation without a separate Higgsfield subscription. Astra handles the research and brief, and ChatGPT's image-generation tool produces the image. Don't treat "Astra" as the name of the image model.

Process:

  1. Attach the real product photograph, the approved brief and relevant brand references.

  2. State each reference's role: product truth, brand style or layout guidance.

  3. Request one image for one approved concept.

  4. Inspect product accuracy before asking for more versions.

  5. Make targeted edits, naming what must stay fixed.

Image prompt:

Generate one Instagram Feed ad image for approved concept [ID].

Reference 1: our real product photograph; use it as the product reference.

Reference 2: our brand style reference.

Approved brand and product brief: [attach/paste]

Audience situation: [specific situation]

Core message: [one approved benefit]

Visual: [describe the scene or product demonstration]

Composition: [product position, focal point, space for copy]

Colors: [approved palette]

Lighting and materials: [specific direction]

Target canvas: portrait 4:5.

Preserve the actual product's shape, construction, color and packaging details.

Do not invent features, accessories, claims, awards or reviews.

Use an original composition.

Exact headline, if included: "[approved headline]"

Place it [position]. Add no other promotional text.

If exact typography is critical, produce the scene without added text instead.

Revision prompt:

Edit the selected image only. Change [specific element] to [requirement].

Preserve the approved product, scene, crop and all other wording.

Do not introduce new objects or claims.

Check the result visually after every edit, because asking the tool to preserve a logo, label or product is not a guarantee. When exact packaging or typography matters, generate the setting only. Then composite your original product cutout, official logo and typeset copy in your design tool, which also makes headline tests easier to control.

Deliverable: one approved image per concept, with the exact copy, prompt and approval status saved alongside it.

Step 5: Higgsfield, Resizing and Pre-Launch Checks

The rest of the SOP covers the Higgsfield route (connecting it to ChatGPT, the Ad Strategist paid-ads workflow with its approval-before-spend prompt, and Ads Studio), placement-specific sizes for Feed, Stories and Reels, and the full pre-launch checklist for product accuracy, claims, rights and Meta ad standards.

The Quick Read:

The Tools List:

📞 Trellus - Power and parallel dialing inside your CRM.

🛠️ Wonderish - Building AI-native tools that turn operators into companies of one.

🤖 AI camp - Access multiple LLMs, assistants, and tools, for teams of all sizes.

📧 SaneBox - Read the important emails in your inbox.

ABOUT THE WRITER:

Meet Joanna (Jo) Lambadjieva

Jo Lambadjieva is an entrepreneur and AI expert in the e-commerce industry. She is the founder and CEO of Amazing Wave, an agency specializing in AI-driven solutions for e-commerce businesses. With over 13 years of experience in digital marketing, agency work, and e-commerce, Joanna has established herself as a thought leader in integrating AI technologies for business growth.

For Team and Agency AI training book an intro call here.

background