Debug Convert in Chrome DevTools

What Convert traffic looks like

Convert spreads across more hostnames than most testing tools, across two domains, which is the first thing to get straight. The split is not arbitrary: the CDN hosts deliver the experiment configuration, and separate hosts collect what happened.

Endpoint pattern

Experiment configuration is delivered from cdn.convert.com and the numbered cdn-3.convertexperiments.com and cdn-4.convertexperiments.com hosts; results are collected at logs.convert.com, logs.convertexperiments.com, and the metrics. and signals. hosts on convertexperiments.com

Transport

Script load for the configuration bundle, then collection calls for visits, experiment exposures, and goal conversions

Client API

The _conv_q queue for deferred calls, plus convert. helpers exposed once the bundle has initialized

Cookies

The _conv_* cookies hold the visitor identifier and the record of which experiments and variations the visitor has been assigned to, which is why a returning visitor stays in the same variation

The two halves, and what each one means when it appears:

cdn.convert.com / cdn-3 / cdn-4 …    experiment configuration delivered
logs. / metrics. / signals. …        visits, exposures, and goal conversions

Debugging Convert with Event Watcher

• No dedicated parser. Requests are matched by hostname, named Convert, badged with the A/B Testing category, and shown with parsed query and body parameters plus cookies set and sent.
• Group by Endpoint first. Convert's several hostnames make the raw stream look busier than it is, and the grouping collapses it into the two questions that matter: did the configuration arrive, and did anything get logged.
• Configuration without collection is the signature failure. The experiment applies to the page, the visitor sees the variation, and the report stays empty because no exposure was logged.
• The _conv_* cookies in the Cookies section explain returning-visitor behaviour. The stored assignment is why someone who should be re-bucketed is not.
• Script Tree shows whether the bundle came from the page's own markup or a tag manager. Convert markets itself on flicker-free delivery, and that depends entirely on the bundle arriving before the page renders.

Convert and consent

Event Watcher checks Convert against the functional consent category.

Convert positions itself as the privacy-conscious option in this category, and its configuration supports that: collection can be held until consent resolves rather than fired on load. Positioning is not verification, though, and the two are easy to conflate on a platform whose marketing does the reasoning for you. The check is the same as for any testing tool: walk the page with the CMP undecided and confirm the collection hosts stay silent.

Common debugging scenarios

• An experiment reports no traffic. Look for the exposure call on a collection host. Configuration requests alone do not populate a report.
• A returning visitor never re-buckets. Read the _conv_* cookies. The assignment is stored there.
• Flicker before the variation appears. Check where the configuration bundle sits in the load order relative to the page's own rendering.
• Goals not converting. Confirm the goal call fires at the interaction and carries the expected identifier.
• Convert fires before consent. Filter to Convert and compare the timing of the collection hosts against the CMP decision, the same comparison worth running against AB Tasty or any second testing tool on the page.