Research question 02 · Evidence allocation

What evidence does ChatGPT need to recommend a smaller Shopify brand?

There is no public recommendation checklist. A small-dog harness sample shows how to decide which product facts, demonstrations, reviews, and boundaries deserve the next 30 days of budget.

Public research sample: KorrikoObserved July 23, 20268 min read

ChatGPT does not publish a checklist that says: add these five pieces of evidence and your product will be recommended. The practical question is: what can a buyer, reviewer, and recommendation system verify for one buying situation—and which missing proof is worth funding now?

Research boundary

Korriko is a public research sample, not a client result. The dated ChatGPT and Gemini observations do not reveal causality, stable rankings, or a fixed recommendation formula.

Start with one buying situation, not the whole category

“What is the best dog harness?” collapses comfort, pulling, escape prevention, step-in fitting, hiking, and car restraint into one list. The narrower research job was an 8–15 lb dog that hates anything going over her head but still pulls during everyday walks.

Korriko’s public product information documented step-in fitting, a five-second fitting claim, a front no-pull D-ring, breathable padded construction, and small-dog sizing. Those facts made the product eligible for the job. They did not automatically make it recommendable. One ChatGPT answer and one Gemini answer sampled on July 23 did not include Korriko; that 0/2 sample only exposed the next evidence decision.

Evidence layer 01

Verifiable product facts

A buyer should be able to verify how the dog enters the harness, leash attachment points, sizing, materials, adjustment points, intended use, and safety limits across the product page, FAQ, retailer listings, and creator brief.

Budget decision: maintain and clarify. Make existing facts specific, consistent, and comparison-ready. A longer description alone is not a recommendation strategy.

Evidence layer 02

Demonstrate the complete buying job

A useful demonstration should show a small dog resisting over-the-head fitting, the complete step-in process, adjusted fit, front attachment during an ordinary walk, and what changes when the dog pulls. That is more useful than another polished unboxing.

Budget decision: test. Fund a small number of task-specific demonstrations before a broad creator campaign. If the complete job cannot be shown cleanly, distribution will only amplify uncertainty.

Evidence layer 03

Independent corroboration

Useful reviews should say whether the dog accepted fitting, sizing was accurate, the harness rotated or rubbed, the front attachment behaved as intended, and which dog shapes or behaviours made it a poor fit. Generic praise supports trust but does not close the decision gap.

Budget decision: brief narrowly. Give reviewers the real job, measurements, and observation questions. Do not script a verdict or buy undifferentiated mentions.

Evidence layer 04

Credible boundaries

The adjacent “escape-proof harness” scene required third belly straps, multi-point adjustment, and escape-specific retention structures in the observed recommendations. Korriko’s public product information did not document those structures.

Budget decision: do not fund. Publishing “escape-proof” content would make the claim stronger than the evidence. A truthful “not designed as an escape-proof harness” boundary is more credible than an ambitious comparison page.

A practical evidence ladder

  1. Define the exact buying situation.A broad category does not provide a usable evidence target.
  2. Identify the public facts that make the product eligible.Use current, comparison-ready evidence—not proposed messaging.
  3. Demonstrate the complete job.Separate attribute claims from observable use.
  4. Ask an independent source to test the important claim.Request conditions, measurements, limitations, and observations.
  5. Define the evidence that should stop investment.Do not turn missing product structure or safety support into a content project.

What the next 30 days should produce

For the step-in harness scene, a rational plan would create clearer small-dog fit evidence, one complete fitting-and-walk demonstration, a small number of independent tests using the same observation criteria, consistent facts across first- and third-party surfaces, and an explicit decision not to fund the escape-proof scene.

The useful output is a budget map: which claim needs clarification, which job needs testing, which outside proof is worth pursuing, and which scene should receive $0.

Continue the method

First choose the buying scene before investing in its evidence.

Read the next decision page