Define the visit before calculating growth
Decide which sources and parameters count in your report. Document your analytics rules and preserve the raw source where possible. OpenAI states that ChatGPT search referrals include a source parameter; other experiences and browser behavior can differ. Inspect actual incoming data before building a channel grouping from assumed names.
Keep internal testing out of the commercial interpretation. A founder clicking a newly published link can create a valid referral event without representing a buyer. Small samples are especially sensitive to these visits. Show counts alongside percentages so a dramatic growth rate does not hide a change from one visit to two.
Connect landing pages to buyer intent
Group visits by the task the page serves: learning a category, assessing a use case, comparing options or contacting the team. A technical guide may attract a useful audience without creating immediate enquiries. A comparison page should help readers understand fit and find the appropriate next step.
Inspect the page after the click. Does it answer the question reflected in the referring source? Does the primary action work on mobile? Is a product condition clearly stated? These checks explain more than a single channel total. Improve the experience where the reader lands before assuming the next priority is more traffic.
Separate three types of evidence
The first type is observed referral traffic: a visit with identifying source information. The second is self-reported discovery: a prospect says an assistant helped them find or evaluate you. The third is sampled visibility: your brand or page appears in an answer collected by your monitoring tool.
Report these side by side, but do not add them as if they were unique people. A single buyer could appear in the first two categories, while the third contains no visitor at all. Where the journey is uncertain, retain the uncertainty. This keeps reporting useful for decisions without turning incomplete attribution into invented precision.
Evaluate quality, not only session volume
Use your existing qualification criteria: relevant company type, appropriate problem, feasible scope and a credible next step. Track how many enquiries meet those criteria and what happens in the sales process. Keep personally identifying information in the appropriate customer system rather than a general content spreadsheet.
Compare AI referrals with other channels using the same definitions and comparable periods. Avoid declaring a channel superior from a handful of conversions. A small number of highly relevant conversations can justify further investigation, but the decision should reflect the sales context and workload needed to support those visitors.
Make the report change an action
A weekly report can show identified visits, landing pages, qualified enquiries, sampled answer observations and changes published. Explain tracking limitations and shifts in the sample. Include one or two concrete next steps, such as correcting a cited pricing detail or improving the page that attracts the most relevant comparison visits.
If traffic is absent, inspect whether the site is discoverable and whether your question sample reflects actual customers. Do not create a new attribution story merely to justify activity. The purpose is to learn which information and channels help buyers move forward, then allocate production effort accordingly. Some weeks the right conclusion will be that more observation is needed.