Discussion starter: an AI search question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.
- AI Search
Agentic browsers like ChatGPT Atlas and Comet act on pages for users. Can you identify agent sessions in your logs, and are they converting?
Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.
1 reply
-
Niraj Raut
Identifying agentic browser sessions reliably is hard in 2026, and I would be cautious about any tool that claims to do it precisely. Browsers that act on pages for users, such as ChatGPT Atlas and Perplexity’s Comet, are built on standard browser engines, and many sessions look like ordinary browser traffic in logs. Some agents identify themselves or arrive with recognisable referrers, and some behave differently from people, moving through pages and forms faster and more directly. Combining those signals gives a partial picture. On conversion, public evidence is thin and volumes appear small for most sites. The more useful work right now is making sure agents can complete tasks on your site when they do arrive, and setting up measurement so you notice when the share grows.
Signals that can help identify agent sessions
Signal Reliability User agent or client hints that identify an agent Good when present; check each vendor’s documentation, since many present as ordinary browsers Referrers or parameters from the agent’s platform Useful when present; not consistent Behaviour: very fast, direct navigation; form fills without typing patterns Suggestive, with false positives Bot management classifications from your CDN Varies by provider; review how they categorise agents Measuring what you can
- Segment known agent traffic where the user agent or referrer identifies it, and track sessions, paths and conversions.
- Monitor behavioural anomalies such as sessions completing checkout unusually quickly, and review a sample to see whether they look like agents acting for real customers or like fraud.
- Check your bot management reports to see how your CDN classifies these sessions, and make sure legitimate agents are not blocked by default.
- Ask customers. A checkout or post-purchase question about how they shopped can capture agent use that analytics misses.
Making your site usable by agents
Agents operate pages much as assistive technology does, relying on the structure of the page rather than visual appearance. Chrome’s Lighthouse added an experimental “Agentic Browsing” category in version 13.3 that looks at the accessibility tree, form annotations and layout stability, and Google marks it as under development. The practical steps overlap with accessibility work:
- Real buttons and links with clear accessible names.
- Form fields with labels, sensible input types and clear error messages.
- Stable layouts that do not shift as elements load.
- Checkout flows without unnecessary steps, pop-ups or interruptions.
- Clear product, price, availability and policy information in text.
Security and fraud
Agentic browsing blurs the line between automated traffic and customers. Bot protection that blocks all automated behaviour may block legitimate agents acting for real buyers; protection that allows all of it may let fraud through. Review how your bot management treats agent traffic, and consider risk-based controls at sensitive points, such as payment and account creation, rather than blanket blocks across the site.
Hypothetical example:
An online retailer notices a small number of orders completed in under a minute from product page to confirmation, from sessions using normal browser user agents. Reviewing them, it finds most are genuine customers whose orders match their account history, consistent with agent use. Its bot protection had been challenging some of these sessions at checkout, causing failed orders. The retailer adjusts rules to apply challenges based on payment risk rather than speed alone, fixes unlabelled form fields in checkout, and adds a post-purchase question about whether an AI assistant helped. Agent-assisted orders remain a small share, but failed orders from that group fall.
Watch for this:
Treating every fast, automated-looking session as a bot can block customers whose assistants shop for them. Treating every such session as a customer can expose you to fraud. Base decisions on risk at sensitive steps, not on a single behavioural signal.
Are agents converting?
For most sites, agent-driven sessions are still a small share of traffic, and there is little public data on how they convert. Where agents complete tasks for users who have already decided, conversion per session can be high; where they browse and compare, it may be low. Build the measurement now, however imperfect, so you have a baseline when volumes grow, and prioritise usability fixes that also help customers using assistive technology.
Segment the agent traffic you can identify, supplement with behaviour, bot management data and customer questions, make pages and checkout usable through clear structure and labels, and adjust security to assess risk rather than block automation outright.
Need help with this on your own site?
Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.
Add a reply Cancel reply
Replies are open to approved members. Log in or apply to join.