How to analyze Scout photos with Claude, Copilot or ChatGPT
You can analyze a Scout photo with Claude, Copilot or ChatGPT by uploading the image together with its tags (store, chain, campaign, date) and a prompt that spells out what good execution looks like in your business, and a field sales manager gets a structured review back in the time it takes to refill a coffee.
Take a typical case: a branded end-cap tower in a supermarket, lit header, six round tiers of multipacks, a cooler on one side and another brand's tower across the aisle, which adds up to a lot of execution detail in a single frame, and the photo probably collected a few likes in the team feed without anyone checking what the middle shelves looked like.
How do you analyze in-store photos with AI?
Open the photo in Scout, save it with its tags, and drop both into whichever AI tool your company has approved. A direct connection between Scout and your AI tools is in development (we're working on an MCP integration so the AI can read Scout and the rest of Salesframe without anyone downloading anything), so for the moment there's some copying involved, though that still beats writing an audit by hand after a long day of visits.
What goes into the prompt decides most of the quality of what comes back. The AI needs the photo, the tags that say where and when it was taken, the standard you judge a store against, and the format you want the answer in. Of those, the standard matters most, because without your perfect store criteria the AI scores against a generic idea of a good store, which won't match what your team was briefed on, so the number looks precise and means very little.
A prompt you can copy for a perfect store review
Replace the bracketed parts with the tags from Scout and your own criteria, and keep the last line, since it stops the AI from filling gaps with confident guesses.
You are reviewing an in-store execution photo for a field sales manager at a beverage company.
Context from Scout tags:
- Store: [store name]
- Chain: [chain]
- Campaign: [campaign name]
- Date: [date]
Scoring criteria (100 points total):
- Display visibility and branding, 25 points: branded header or POS materials present, lit, unobstructed
- Product availability and fill level, 25 points: every tier stocked, no visible gaps
- Facings and planogram compliance, 20 points: packs front-facing, correct brand blocking, no mixed products
- Pricing, 15 points: price labels present and readable
- Adjacent and competing displays, 15 points: what sits next to this display and whether it helps or competes
Tasks:
1. Describe what you can see before scoring anything.
2. Score each criterion with a one-sentence reason, then give the total.
3. List what you could not judge from the photo (back-stock, expiry dates, price accuracy) instead of guessing.
4. Give the rep 2 specific fixes for the next visit.
5. Write a 3-line summary I can paste into the team channel.
Do not score anything you cannot see.What the answer looks like
Run against the end-cap described above, with that sample rubric, a reasonable response reads something like this (the numbers are illustrative, and your own rubric will produce its own):
Display visibility and branding, 23/25. Lit header and branded halo are clear and unobstructed, and the same brand colors carry through to the shelving in the neighboring aisle.
Product availability and fill level, 15/25. The top and bottom tiers look full, while tiers two to four are visibly thinner, with gaps at the front.
Facings and planogram compliance, 17/20. Multipacks face forward with the brand label showing, and no mixed products are visible on the tower.
Pricing, 6/15. One price label is readable on the second tier, and none can be confirmed on the others.
Adjacent and competing displays, 10/15. Another brand's tower in the same format sits directly across the aisle, so the display shares attention instead of owning the end of the aisle.
Total: 71/100. Fixes for the rep: refill the middle tiers, add price labels to each tier.
The part worth reading closely is the list of things the AI said it could not judge, because that's where a human still has to decide: back-stock behind the tower, whether the price shown matches the promotion, and whether the tower stands where the planogram says it should.
Where AI gets it wrong
Chat-based AI tools misread small text, miscount packs when the angle hides a row, and can't see anything behind or above what the camera caught. They're also consistent without being calibrated, so the same photo might come back as 71 one time and 68 the next. Treat the score as a way to sort photos into "look at this one first" and "fine," keep the rubric identical across every photo you run, and ask the tool to flag uncertainty rather than hide it.
For a manager reviewing a few dozen photos from a campaign, this works well. For auditing thousands of outlets with SKU-level share of shelf and automated planogram compliance, dedicated image recognition platforms exist for that job, and a chat window isn't the right tool for it.
From one photo to a whole region
Single photos are a decent demo, though the bigger payoff comes when you drop in a batch with the tags attached, because Scout tags every photo with store, chain and campaign at capture, and that structure is what lets the AI compare like with like. Try a prompt along these lines:
Here are [number] Scout photos from the [campaign name] campaign, each with store, chain and date tags. Score each against the criteria above, then tell me: which chain is weakest on pricing, which 5 stores I should have a rep revisit first, and whether end-cap execution differs between chains. Flag any photo you could not score and say why.
Photos that sit in a camera roll or a group chat with no tags can't be compared this way, since the AI has no way to know which chain or campaign a given shelf belongs to, which is the whole reason structured capture matters before any analysis does.
Before you upload store photos
Use the AI tool your IT team has approved, since store photos belong to the company and some contain shoppers or store staff, so crop or skip those frames, and check your policy on uploading images to external tools.
FAQ
Can ChatGPT, Claude or Copilot score a perfect store from a photo? Yes, as a first-pass review, provided you give the tool your own perfect store criteria and weights. Without them it invents a generic standard, and a human should still verify anything the tool says it could not see.
Does Scout connect directly to Claude, Copilot or ChatGPT? Not yet. An integration that lets your AI tools read Scout and Salesframe directly is in development. Until then, you save the photo and its tags from Scout and drop them into the AI tool manually.
Which AI tool is best for analyzing store photos? All three read images well enough for this job, so the better choice is whichever your company has approved. The quality of your criteria and the consistency of your prompt matter more than the model.
How many photos can you analyze at once? That depends on the tool and your plan, though a batch of a few dozen photos from one campaign or chain is a reasonable size. Beyond that, split the batch by chain or region so the comparison stays readable.