Check Your Direct Traffic Before Crediting AI Search

Kevin Indig’s October 19 analysis of 12 websites gives marketers a reason to inspect a traffic spike before counting it as an AI win. Your next budget report needs evidence about the visitors behind that increase.

Across his filtered January–September sample, total Direct sessions rose 140.9%. Average daily engaged Direct sessions rose 2.3%. The comparison adjusts for different month lengths; monthly engaged counts fell 1.0%. These are different measures of the same sample, rather than competing estimates of one growth rate.

The organic growth advisor and Growth Memo author also tested four brands separately for relationships between AI visibility and subsequent traffic. After checking the apparent correlations, Indig wrote:

“An effect could still exist. I just couldn’t isolate it.”

That uncertainty belongs beside a request to increase the AI-search budget. An unexplained visit doesn’t establish which marketing activity earned it.

Start with what Direct actually records

Google’s own definition is traffic without a clear referral source. A person typing a web address can appear there, but so can a visit whose tracking information disappeared. Google identifies missing campaign tags and redirects among the causes. Links in offline documents and interference from ad blockers also complicate the record.

For a small marketing team, that creates a practical order of work. Check the tracking on your existing campaigns before creating a new explanation for the unassigned visits. Follow an email link through its redirects and inspect the address that arrives at the website. A missing tag is a specific repair you can make. Leaving it unresolved makes the next channel comparison less useful.

Treat Direct as a place to investigate. Changing its label to AI search would preserve the uncertainty while making the report sound more confident.

Check what the visitors did

Indig’s sample skewed toward B2B by site count and mostly targeted the US. Traffic volume was almost evenly split between B2B and B2C, a different balance from the site count. He excluded common direct-entry pages and identified suspicious groups of visits through browser and engagement patterns. Those judgments aren’t an independently verified identity for every session.

Google’s engagement documentation supplies another useful limit. An engaged session lasts longer than 10 seconds, includes at least two page views, or triggers a key event. Passing that threshold doesn’t demonstrate that the visitor is a potential customer. It gives the analyst an additional behavior to inspect.

Keep the total and engaged figures visible together. Then look at the landing pages behind an increase and whether the same visits produce meaningful business activity. A traffic chart that rises while qualified inquiries stay unchanged warrants a different budget discussion from one supported by new customers.

Check the measurement setup too. Google warns that artificial background pings can inflate engagement. A team that changes tracking should record that change beside the reporting period; otherwise, an instrumentation improvement can be mistaken for a marketing improvement.

Our coverage of Bing’s topic-planning report makes a related distinction: an AI citation can guide an investigation without proving demand. A visit with no identified source leaves a different gap in the same reporting chain.

Ask the customer before assigning the credit

The four-brand exercise tested 632 relationships, and the apparent results didn’t remain reliable across the author’s timing and split-period checks. It doesn’t settle the question for another business. Indig sells advice in this field, and the article includes a sponsor.

His recommendation includes a direct request:

“Ask new leads and customers where they first heard about you”

Use the response as additional evidence. A customer can identify an AI tool they consulted and explain the question they were trying to answer. Keep their wording rather than automatically mapping every mention into a channel total. A recalled recommendation and a measured referral describe different parts of discovery.

This also connects attribution to the quality of the content AI recommends. Learning which question brought a prospect to you gives the content team something more specific to examine than an unexplained increase in visits.

Before the next marketing budget review, put confirmed AI referrals beside customer-reported AI discovery. Keep unexplained Direct visits separate. Ask the analyst to show where those groups overlap and which led to qualified inquiries. Approve the next experiment against an outcome the team can actually check.