Claude
Gemini
ChatGPTNova surfaces every Amazon fee, refund, and margin shift in your live P&L, across 23 marketplaces. See it in your data
What happened
India's three largest marketplaces have moved their AI budgets from shopper-facing features to seller-facing ones. Amazon India, Flipkart and Meesho are all now competing on listing generation, catalogue creation, pricing help and automated ad targeting rather than on chat assistants for buyers (Inc42, August 18, 2026).
The numbers being quoted are operational, not promotional. Amazon says sellers using its tooling cut listing errors by 10% and routine operational work by nearly 70%. Meesho's AI voice system handles up to 300,000 seller calls a day, and its annual transacting seller base reached 1.04 million, up 81% year on year.
Why marketplaces are spending here
Buyer-side AI is expensive and hard to attribute. Seller-side AI is cheap by comparison and pays back twice: it removes support cost, and it increases the number of sellers who can list, price and advertise without help. When a marketplace's growth depends on supply, making supply cheaper to onboard is the highest-leverage thing it can fund.
The second reason is quieter. Every one of these tools sits inside the marketplace. The seller who lets the platform write the listing, set the price band and pick the targeting has handed the platform three of the four levers that decide their margin.
The tension nobody in the announcements mentions
Meesho's CTO described ranking as a balance between relevance, fairness, new-seller inclusion, revenue and quality. Those five goals do not point the same way. Revenue pulls toward whoever pays most, new-seller inclusion pulls toward whoever pays least, and the seller sitting in the middle has no visibility into which weighting applied to their listing this week.
That is the case for keeping your own numbers outside the platform that produces them. A marketplace tool tells you your listing improved. It does not tell you whether your contribution margin improved.
What this means outside India
India is where Amazon tests seller tooling at volume and low cost before it appears elsewhere. Listing generation, automated targeting and voice-based seller support all follow the same route the Seller Assistant took. Expect the same features in mature marketplaces with a different name and a slower rollout, and expect the same open question: the tool optimises for the metric the platform chose, which is rarely your profit per unit.
What to do about it
- Use the free tooling, keep the scorecard separate. Let the platform draft the listing. Judge the result on your own margin numbers.
- Log every automated change. If a tool rewrites titles or bullets, snapshot the before state so you can attribute a conversion shift later.
- Watch ad spend per unit, not ROAS. Automated targeting looks good on ROAS long after it has stopped being profitable per unit.
- Point your own AI assistant at your own data. An assistant that reads your live P&L answers a different class of question than one that reads the marketplace's view of you.
How Nova helps
- Nova MCP - connect your Amazon data to Claude, ChatGPT or Gemini and ask questions against your numbers instead of the marketplace's dashboard.
- Listing Health Scanner - independent check on listing content, so an automated rewrite gets audited rather than trusted.
- PPC Analytics - product-level ad spend against profit, which is the number automated targeting does not optimise for.