Short answer
With ChatGPT's new wallet, AI agents now find the product and complete the payment too. According to Uğur Deniz İlhan, co-founder of UNIT İstanbul, this means the decision is moving from the brand's website to the data the agent reads; brands need to fix their product data, AI visibility and bot rules now.
Highlights
- The agent does not look at the shop window, it reads data: price, stock, delivery and review information is where decisions now happen.
- Three risks for brands: not appearing in AI answers, being misread, and blocking agents without realising it.
- When building agents in-house: suggestion mode first, human approval and a log of every step.

3 min readAuthor: UNIT Journal EditorEditor-in-chief: Uğur Deniz İlhan
AI assistants are moving from making recommendations to finishing the job. According to TechCrunch, with the Wallet feature OpenAI has added to ChatGPT, the agent finds the product, fills the basket and completes the payment once the user approves (details). For now the feature is open to a limited group of users in the US, but the direction is clear.
Where do decisions happen in the age of agents?
According to Uğur Deniz İlhan, co-founder of UNIT İstanbul, the biggest change is that users can now decide without ever visiting the website: "The user tells their agent 'find it, compare, buy it'. The agent does not look at the shop window, it reads data. The brand's website still matters, but the decision is increasingly made on the price, stock, delivery and review information the agent reads."
Three risks for brands
- Not appearing: if the agent never sees the brand in AI answers, the brand stays out of its recommendations.
- Being misread: if price, stock or delivery information on the site contradicts the structured data, the agent decides on wrong information.
- Blocking without realising: generic bot defences can shut agents out too; this should be a deliberate decision, not an accident.
What should be done today?
- Clean up product and service data: price, stock, variants, delivery and return terms should match and be current on the page and in structured data.
- Regularly check that AI answers describe your brand correctly; GEO is no longer just about visibility, it is part of the path to a sale.
- Track visits from agents and AI assistants as a separate channel in analytics.
- Review your bot and security rules; know which agents you allow and which you block.
For companies using agents in-house
Uğur Deniz İlhan recommends the same discipline when brands build their own agents: "Run the agent in suggestion-only mode first. Every step that spends money or goes outside should pass a person's approval, and every step should be logged. Managing agents is as much about responsibility as it is technical work." We explain what the role that carries this responsibility does in What is an AI Agent Manager; for building agents, see AI agent development.
*Uğur Deniz İlhan is co-founder of UNIT İstanbul. As AI Agent Manager and Growth Hacker, he leads growth and AI agent projects for brands in Turkey and internationally.*
Frequently asked
- Can an AI agent go shopping?
- Yes. With ChatGPT's Wallet feature, the agent finds the product, fills the basket and completes the payment once the user approves. For now the feature is open to a limited group of users in the US; similar features are expected to spread.
- How do I know if my brand is ready for AI agents?
- Check whether your product or service information is consistent on the page and in structured data, whether AI answers describe your brand correctly, and whether your bot rules block agents deliberately. These three are the precondition for an agent to find you and read you correctly.
Links
Sources
- TechCrunch ·
Follow UNIT Journal
What's new in search, AI and technology, in your feed every day.


