How the AI pricing works (and where it falls short)

We built Clutter to Cash because I spent too many evenings pricing used goods by hand, and it seemed silly that every person selling a camera lens or a pair of jeans had to re-do the same research from scratch. This guide is the honest, technical version of how the tool works under the hood: what data we use, what the AI is actually doing, and the specific situations where the tool gets things wrong. If you're going to trust a price we give you, you should know where it came from.

What happens when you hit "Analyze"

When you upload photos and submit, three things happen in sequence. We'll walk through each one in turn.

1. Identification. Your photos are sent to a vision-capable AI model (Google's Gemini Flash is our first pass). The model is asked what the item is, what brand and model (if visible), and what condition the photos suggest. It returns a structured answer with a confidence score.
2. Pricing research. Once we have a confident identification, we search live marketplace listings that match it, and bring in a second source when the first search comes back thin.
3. Listing generation. A text model uses the identification, condition notes, and pricing data to draft a title and description for the marketplace you picked (eBay, Facebook Marketplace or Mercari), next to the price range. You see the draft. You copy it. You leave.

Stage 1: how identification actually works

The vision model sees your photos the way a human would and can often recognize a specific model number, label, or brand mark from a photo without you having to name it. When the photo is clear and the item is mainstream, Gemini Flash can often identify it with confidence on the first pass.

For the rest (blurry photos, missing labels, obscure items, counterfeit goods) we fall back to a two-step enrichment. First we run the primary photo through Google Lens to pull visually similar products from the open web. Then we re-ask the model with that extra context. If the confidence score is still below threshold, we ask Gemini Pro, a slower and more expensive model, for a second opinion. The most confident answer wins.

The key design decision here is the confidence threshold itself. If we ship a confident answer on a low-quality signal, we send you a wrong price and waste your time. If we threshold too conservatively, we reject too many legitimate queries. The threshold sits on the model's self-reported confidence. Below it, the tool pauses before pricing and puts the choice in your hands: pick from its best guesses, type what the item is, add more photos, or skip ahead to a price search. Above it, the tool still asks "Is this the [item]?" with a Yes and a Fix it, so a wrong identification gets caught before it becomes a wrong price.

Stage 2: where the price numbers actually come from

The price comes from real listings, not from the AI. The anchor today is a live search of currently-listed items matching the identified model and category, run fresh at request time — not a cached, aging database. When that first search comes back thin, we bring in a second live source so the range isn't built on a couple of data points.

Asking prices can run higher than what items actually sell for — an asking price is what the seller hoped to get, not what a buyer agreed to pay. So the results page says the range is built from asking prices, and how many. The suggested price starts from the middle of what the search found, so one oddly priced listing doesn't set it, then comes down for the condition and any visible flaws the model spotted (you can correct those). The range around it runs from the lower to the upper end of what was found, so with only a few listings it can be wide.

We don't blend in live data from every marketplace today — broader coverage is on the roadmap, not in the tool yet. For unusual categories where the search isn't a natural fit (vintage furniture, niche collectibles), expect fewer comparable listings. With none at all, the tool says so and shows no price rather than guessing. With only a few, it shows the range with a note to check a listing or two yourself.

Stage 3: listing generation

The final step is the listing copy. A language model takes the identification, your condition notes, and the comp-derived price and drafts a title plus description for the marketplace you picked (eBay, Facebook Marketplace or Mercari), following the structure described in our listings guide. The model is instructed to be keyword-dense in the title, specific about flaws in the description, and to never invent features or condition claims that weren't in the input.

You should treat the generated listing as a strong first draft, not a finished product. Edit it. Add context the AI couldn't know: where you bought it, why you're selling, how long you've had it. The AI is explicitly instructed not to fabricate these details, because getting them wrong is worse than leaving them blank.

Where the tool gets it wrong

Being honest about the failure modes matters more than hiding them. Here are the specific situations where you should not trust our output as-is.

• Luxury authentication. The AI cannot tell a genuine Louis Vuitton bag from a good counterfeit. For items where authenticity materially changes the price (designer bags, high-end watches, collectible sneakers) use a dedicated authentication service before you list.
• Items with no meaningful comp history. One-of-a-kind vintage pieces, handmade goods, regional or discontinued products, and brand-new releases often don't have enough listings to anchor a price. The results show how many prices the range was built from; when that number is small, treat the range as rough.
• Local market variance. A snowblower in Buffalo is worth more than a snowblower in San Diego. We don't adjust for your location. If the item is heavy enough that shipping is prohibitive and sales are local-only, our national-average price may overstate what you can actually get.
• Ambiguous identifications. The AI sometimes picks the more famous variant of a product when the real answer is the less famous one. Two generations of the same phone can look nearly identical in a photo, but their prices are very different. Always spot-check the year and exact model against your item.

Data practices

Your photos are uploaded as you send them to Cloudflare R2 under an anonymous, session-scoped path, and handed to the AI providers for analysis. Stored 6 months, inputs and outputs, only to improve accuracy — then deleted. Never sold. Never shared with anyone beyond the AI providers needed to run the tool — and they never get your name, email, or location, because we never asked for it. Deletion is automatic, by an R2 lifecycle rule. During the retention window the inputs are used only to improve the tool itself: better identification, better pricing, better drafts. They're never tied to a person (no account exists to tie them to). Our full data practices are documented in the privacy policy, and the AI providers we use — Google Gemini for identification (with a Google Lens lookup through SearchAPI.io as fallback) and for the listing draft, plus SerpApi and Serper.dev for price research — are disclosed in the disclosures page.

That's the whole pipeline. If you've got questions about a specific result the tool gave you (or a case where you think we got it wrong) write to info@icandothat.ai. If a price looks off, the results page has a link to tell us, and feedback on mispriced categories helps us decide what to improve.