Procurement framework
Five criteria for evaluating an agent commerce platform
Coverage, answer evidence, source-level fixes, purchase-path testing, and source tracking determine whether a platform is fit for day-to-day operations.
Define the buying category
Acquisition tools optimize. GEO monitors. Only a channel sells.
Tools cut costs or watch the answers. A channel closes the sale.
AI-assisted acquisition improves the existing funnel; agent commerce turns AI into a sales channel. SEO, creative, ad copy, and campaign tools cut costs inside existing search, social, and ad channels. GEO (Generative Engine Optimization) tools measure how brands appear in AI answers.
AI acquisition
- Primary object
- Search rankings, creative, copy, service interactions, and campaigns
- Business change
- Improve discovery and lower content and execution costs in existing channels
- Distribution surface
- Search, marketplace, social, and ad systems
Produce campaign assets more cheaply, then bid them into the same rising auction. Traffic allocation stays with the platform.
GEO
- Primary object
- Mentions, citations, and recommendations in AI answers
- Business change
- Measure brand presence in AI answers
- Distribution surface
- AI answer interfaces
You see whether you were mentioned. The source stays untouched, and the sale stays out of reach.
Agent commerce (SoldByAI)
- Primary object
- Catalog data, engine results, purchase paths, and source records
- Business change
- Manage AI as a sales channel
- Distribution surface
- Agent shortlists and purchase paths
A machine-readable catalog, verifiable trust, and orderable endpoints: a new distribution surface.
SoldByAI manages this operating chain, from catalog connection to order attribution.
Put these five questions to all three categories. The answers sort themselves.
| Five criteria × three categories | AI acquisition | GEO | SoldByAI |
|---|---|---|---|
| Coverage | Campaign tools do not test AI answers. | Monitoring coverage depends on each vendor's list. | Run records are kept line by line. |
| Answer evidence | Campaign assets produce no observations. | Score reports cannot answer why. | Original answers are kept line by line. |
| Fixes and retesting | Creative can be re-run; the product data stays as it was. | Reports hand the fixing back to you. | Fixes map to products, with your approval and a retest. |
| Purchase path | A click lands in the store; the path itself is never tested. | Monitoring stops at the answer. | Four gates are tested one by one. |
| Source tracking | Ad attribution follows each platform's own rules. | Monitoring never enters your storefront data. | One marker runs from visit to order. |
01 / Coverage
Where is the run record for each listed engine?
Coverage starts with what actually ran
Coverage becomes auditable when the engine list, real buyer questions, test times, and run status remain on record.
A missing measurement is not a measured absence.
AI acquisitionCampaign tools do not test AI answers.
GEOMonitoring coverage depends on each vendor's list.
SoldByAIRun records are kept line by line.
Every major AI shopping entry point included · no per-engine upcharge
- ChatGPT
- Gemini
- Copilot
- Perplexity
- Claude
- Doubao
- Qwen
- Kimi
- DeepSeek
02 / Answer evidence
Which original answer supports the conclusion?
The original answer comes before the score
Mentions, citations, recommendations, and measured absences are recorded separately. Each record keeps the question, answer excerpt, rationale, and test time. Aggregate results stay traceable.
Scores summarize. Original answers supply the evidence.
AI acquisitionCampaign assets produce no observations.
GEOScore reports cannot answer why.
SoldByAIOriginal answers are kept line by line.
- Question
- Real buyer question
- Answer
- Original answer excerpt
- Assessment
- Evidence level · Rationale
- Time
- Tested at
03 / Fixes and retesting
Which brand facts and records from prior sweeps and fixes carry into the next cycle?
Finding the issue is only the start of the fix
Each error or gap is tied to a specific product. The proposed fix keeps the original evidence, action path, and retest status. You approve every change, and the next cycle checks the result.
AI acquisitionCreative can be re-run; the product data stays as it was.
GEOReports hand the fixing back to you.
SoldByAIFixes map to products, with your approval and a retest.
Every change requires your approval.
- Product
- Original evidence
- Action path
- Your approval
- Next-cycle retest
Does the next cycle inherit your brand context, or start over from a generic prompt?
Brand facts, sweep history, and fix records carry into the next cycle. Every cycle builds on the last, making each proposed fix more specific to your brand.
04 / Purchase path
What does the platform test after a product is recommended?
A recommendation still needs a live purchase path
A recommendation is only the start. SoldByAI runs an agent through find, compare, and select, then verifies the purchase route is live. Link reachability is tested separately, with the failure point and test time on record.
The task itself determines whether the purchase path holds.
AI acquisitionA click lands in the store; the path itself is never tested.
GEOMonitoring stops at the answer.
SoldByAIFour gates are tested one by one.
- 01Find
- 02Compare
- 03Select
- 04Route live
05 / Source tracking
How does the source marker persist from the visit to the order event?
Attribution depends on an unbroken source marker
Product links carry source parameters across every supported output. Your analytics can identify the output and product behind each visit. The same marker must persist through checkout and order events before the order can be attributed to an engine.
AI acquisitionAd attribution follows each platform's own rules.
GEOMonitoring never enters your storefront data.
SoldByAIOne marker runs from visit to order.
- 01Product output
- 02Product visit
- 03Checkout event
- 04Order event
Procurement checklist
Nine checks for a defensible decision
Procurement decisions depend on capability scope and the evidence behind each result.
- Engine coverage
Engine list, buyer question, test time, and run status
Cross-Engine Sweep Network
- Test method
Real buyer questions and a consistent measurement method
Cross-Engine Sweep Network
- Answer evidence
Original answers and four evidence levels
Cross-Engine Sweep Network
- Catalog structure
Product identifiers, specifications, price, availability, and source
Storefront Direct Read
- Fix process
Product, original evidence, and action path
Fix-at-Source Engine
- Change approval
Approval status for every proposed change
Fix-at-Source Engine
- Follow-up verification
Brand context, fix records, and retest status in the next measurement cycle
Fix-at-Source Engine
- Purchase path
Four gates: find, compare, select, route live, plus failure point and link reachability
Purchase-Path Live Test
- Source tracking
Product output, visit, and the same marker in order events
Engine-Level Attribution
Three controls belong in every procurement decision
Your storefront stays intact
Your URLs, design, and checkout remain unchanged.
You approve every change
Proposed fixes move forward only with your approval.
Your catalog remains portable
Your data remains yours, and the structured catalog stays portable.
Choose the plan that fits your operation
Four plans, defined by testing cadence, shopping-question allowance, and service scope.