How I AI: Nicole Ruiz’s System for Buying High-Quality Goods and Automating Returns with Claude

Nicole Ruiz built a Claude project to stop panic-buying low-quality household goods, especially for kids. In this episode of How I AI, she shows how she turns her own purchasing standards into reusable instructions so Claude can surface durable products, trusted retailers, and warning signs before anything reaches her cart.

Nicole is trying to solve a familiar problem for busy parents: you need something quickly, search results are flooded with ads and questionable brands, and the easiest option is usually the disposable plastic version on Amazon. Her system starts from the opposite direction. She wants products made from natural materials, companies with a long repair or manufacturing history, and items that can survive years of use in a small apartment with kids.

The two workflows in this episode cover both sides of the purchase cycle. The first turns Claude into a research assistant that compares products against Nicole’s own standards instead of generic popularity metrics. The second uses Claude Cowork to find receipts, gather order details, and draft return or refund requests when a product fails earlier than expected. You can follow the full implementation in Automate Product Returns and Refunds Using Claude Cowork. See Automate Product Returns and Refunds Using Claude Cowork. See Build a 'Buy It For Life' AI Shopping Assistant with Claude.

The automation is deliberately narrow. Nicole still reviews the evidence, decides whether a brand actually seems trustworthy, approves outgoing messages, and sends them herself. The value is in reducing repetitive household admin without handing over judgment.

Turn your purchasing standards into a reusable Claude project

Nicole’s Claude project starts with a constraint most shopping tools ignore: she does not want the entire internet. She wants Claude to search through a filtered set of values and vendors first, then expand outward only when necessary.

Write the checklist you already use mentally

The core move was taking an invisible household decision process and writing it down as persistent project instructions. Instead of repeating the same evaluation every time she shops, Nicole encoded the standards once so Claude could apply them repeatedly across categories like baby clothes, kitchen tools, blankets, and shoes.

Her criteria include:

• Longevity and repairability: Favor products designed to last, especially brands that support maintenance, replacement parts, or repairs instead of disposable upgrades.
• Track record: Look for manufacturers with years or decades of consistent craftsmanship instead of relying on aesthetic branding or an "about" page.
• Returns: Prefer companies that clearly stand behind products and make refunds or exchanges straightforward when quality issues appear.
• Marketing skepticism: Treat heavy advertising, influencer campaigns, and polished direct-to-consumer branding as neutral signals, not proof of quality.

She also instructs Claude to watch for modern shopping pitfalls: drop-shipping operations, AI-generated reviews, unclear sourcing, and brands that suddenly changed ownership or quality. Nicole describes these as heuristics, not hard rules. A private-equity acquisition or a flood of influencer posts does not automatically mean a product is bad, but together those signals can justify more scrutiny.

Seed the project with trusted vendors

Nicole imported a long Apple Notes list of stores and brands she already trusted. That gave Claude a curated starting point instead of forcing every search to begin from generic search-engine results.

Some of the retailers she uses, including Boston General Store and Manufactum, already function as human-curated filters for household goods. Claude searches those sources first, surfaces comparable options, and saves her from navigating difficult retail sites manually. Nicole points out that many heritage manufacturers have terrible websites despite making excellent products, which makes AI search unexpectedly useful for smaller or older businesses.

Force every recommendation into the same format

Nicole asks Claude to return the same fields for every product candidate so she can compare options quickly instead of bouncing between tabs and marketing copy:

1. Product name
2. Photo
3. Price
4. Materials
5. Care and maintenance
6. Purchase link
7. Evidence about the brand’s history

That structure keeps the tradeoffs visible. Nicole specifically wants materials and maintenance requirements surfaced early because those details are often buried far below product photos. A blanket that needs constant hand washing or a kitchen tool made mostly of plastic may still be acceptable, but she wants the compromise exposed immediately instead of hidden in fine print.

Use short prompts because the context already exists

Nicole demonstrated the workflow with a surprisingly ordinary request: finding a durable can opener. Her existing project instructions handled the rest.

Because the purchasing logic already lived inside the project, the actual prompt was minimal. She did not need to repeat her preferences about repairability, materials, trusted vendors, or return policies every time she searched.

Help me find a can opener

Claude returned several options, including a NuGen Super Kim can opener sold through Boston General Store and Manufactum. The response included pricing, materials, retailer links, and background on the manufacturer’s history making kitchen tools. Nicole could then continue the conversation by asking follow-up questions about durability concerns or negative reviews instead of restarting the research process from scratch.

She also showed a different use case: evaluating a brand discovered through an ad. Claude advised against adding it to her trusted list and cited a mix of signals including recent private-equity ownership, management controversy, Glassdoor complaints, influencer-heavy marketing, and declining customer reviews after scaling. Nicole’s point was not that any one factor proves poor quality. The value is aggregation. Claude can quickly assemble weak signals from across the web so she can decide whether a brand deserves further trust.

Use Claude Cowork to prepare returns and refund requests

Nicole’s second workflow starts after a purchase disappoints. In her example, a pair of J.Crew pants wore through after only a few months. The annoying part was not deciding to request a refund. It was the scattered administrative work required to actually do it.

She used Claude Cowork to gather the evidence, locate the order information, and draft the customer-service message while she continued doing other things around the house.

Capture the problem immediately

Nicole photographed the damaged pants as soon as she noticed the issue. That small habit matters because household admin often fails at the collection stage, not the writing stage. If the photo never gets taken, the return quietly turns into another unfinished task.

Dictate the request instead of composing it

Rather than sitting down to write an email, Nicole used voice input while moving around the house. She mentions often using this workflow one-handed while carrying or nursing a baby. The point is not perfect prompting. It is reducing the friction required to start.

Her prompt asked Cowork to locate the receipt, extract the order details, and draft a refund request to J.Crew customer service:

I have this pair of pants of Rafa's from the last six months they've already worn through in the butt. I'm looking to return them to J. Crew and specifically get a refund. Can you help me draft an email and specifically start out by finding the receipt for the pants in my email, either from PayPal or J. Crew with the item number and any other details you might need? Please include that context and draft up the request for refund to J. Crew customer service.

Let Cowork gather the scattered order information

With access to connected tools, Cowork handled several repetitive tasks automatically:

1. Search Gmail for matching J.Crew or PayPal receipts.
2. Extract the item number, order number, size, and purchase date from the receipt.
3. Draft a refund request that included those details and referenced the product issue.

Review the draft before sending

The resulting draft included the order details in the subject line and explained why the level of wear felt unreasonable for the price. Claude also surfaced reports from other customers describing similar quality issues with the item. Nicole still reviewed the wording, checked the references, and decided whether the supporting claims were accurate enough to include.

During the demo, Nicole said the process took under two minutes from photo to prepared email. That timing depended on existing integrations and account access, but the larger point held up: the AI removed the search-and-copy work that usually makes small returns feel disproportionately tedious.

A reusable system for household decisions

The interesting part of Nicole’s setup is not really the shopping itself. It is the idea that recurring household decisions already contain stable criteria, and those criteria can become reusable infrastructure. Her Claude project stores purchasing standards once instead of forcing her to mentally reconstruct them every time she needs toddler shoes, a blanket, or a kitchen tool.

The workflow also creates a path toward smaller retailers and manufacturers that are hard to discover through conventional search. Claude can navigate inconsistent websites, compare products across stores, and surface background information that would normally require multiple tabs and a lot of patience. But Nicole repeatedly keeps humans in the approval loop because seller legitimacy, materials claims, and long-term quality still require judgment.

What is worth copying here is the structure, not necessarily Nicole’s exact taste in products. Pick one repeated household decision that already has a hidden checklist behind it. Encode the standards, require side-by-side evidence, and make the AI show its reasoning instead of just producing recommendations. The return workflow is especially practical for anyone buried in small administrative tasks, but it works best when the AI gathers context while a human still controls purchases, account access, and outgoing communication.