AI reporting: what your data can and can’t tell you

The temptation is to put everything under one heading: AI performance. But a rise in crawler activity does not establish that more people are discovering your business. A longer visit does not prove that an assistant influenced a purchase. Even a sharp movement in a ranking chart may need checking before it warrants changing a page.

The practical challenge is to understand what each source can establish without losing sight of the commercial question: where should you focus your effort?

Start by separating the evidence

The attribution problem provides the background. Similarweb’s downstream-impact research, published on 23 June, found that users were, on average, 2.5 times more likely to visit an AI-recommended brand than a competing brand within seven days. Its sample covered US desktop users researching finance, travel and beauty. Organic growth advisor Kevin Indig’s 19 October analysis could not establish a reliable relationship between AI visibility and subsequent branded search or direct traffic across four brands. [1][2]

Those studies ask different questions using different methods. They give us reasons to investigate the journey, without providing a formula for attributing a business’s enquiries to AI.

For reporting, I would keep four observations distinct:

• AI visibility: whether a business appears in the answers being sampled.
• Crawler activity: requests recorded against content on the website.
• Referral visits: people arriving through an identifiable AI source.
• Subsequent behaviour: what visitors do on the site, including the actions that matter commercially.

Branded search can add useful context, but a rise in searches for a business’s name does not identify what prompted them. These measures can inform one another without being treated as interchangeable.

Use server logs to investigate access, not demand

Log analysis is a useful example of why those boundaries matter.

A request in a log does not establish that the page appeared in an answer, that a person saw a recommendation or that revenue followed. Counting that activity alongside customer visits would obscure the difference.

The useful questions are more specific. Which resources were requested? Are those pages current? Does the pattern justify investigating content that has received little attention?

That can lead to worthwhile work. It gives us a reason to inspect particular content, while leaving the question of customer demand open.

Examine what visitors do after they arrive

An identifiable AI referral gets us closer to a human journey. It still leaves plenty to understand.

Traffic totals can show that people arrived. Engagement and onward navigation help us investigate what happened next. James’s emphasis is on identifying the pages where visitors spend time and the next pages they choose, alongside the traffic and conversion measures already being examined.

For me, that means starting with the landing page and the need it appears to serve. Someone arriving on a homepage might be comparing suppliers, looking for a specific service or trying to access an existing account. Those visits should not automatically be interpreted as the same commercial opportunity.

The investigation should connect three questions: where did visitors arrive, what did they engage with and could they find an appropriate next step?

If relevant information is difficult to find, the next action may be to improve the page or the route onwards. A smaller audience spending time with useful content may also warrant closer attention than its traffic volume initially suggests. Longer engagement alone, however, is not proof of satisfaction or a sale.

Available Google query data can help identify needs worth investigating. It cannot be relabelled as the original prompts people entered into ChatGPT. A landing page and a plausible question do not reconstruct an individual’s research history.

Check what the channel label actually describes

The route to the website deserves the same scrutiny as the visit itself.

Someone who encounters a brand in AI may later search for it and arrive through an advert or an organic result. The recorded channel describes that arrival; it does not automatically explain the earlier discovery.

That is something to check in the actual reporting setup. A broad source label naming an AI platform should not be assumed to distinguish an advert from an unpaid recommendation. Nor should the presence of a branded visit be taken as confirmation that AI created it.

The practical aim is to establish what the label records and what remains unknown before comparing channels or assigning credit.

Validate a ranking movement before reacting to it

Sometimes the uncertainty lies in the data collection itself.

Search Engine Journal’s 23 October investigation compared providers’ accounts of how they handle suspect Google results. SEO platform SEOmonitor’s documentation says some incorrect results entered its platform between 9 and 13 October, before affected keywords were reprocessed on 13 October. It distinguishes those collection problems from genuine ranking volatility. [3][4]

That creates a clear decision point. Before rewriting a page in response to a sudden reported drop, check which queries and pages are involved, what changed in Search Console and analytics, and whether the provider has issued a relevant notice.

My preference is to begin with Search Console for the site’s actual Google search performance.

The sources serve different purposes. A tracker can guide an investigation without being sufficient evidence for a major change. Google also states that AI Overviews and AI Mode traffic is included in Search Console’s overall Web performance data; that does not identify every organic visit as AI-influenced. [5]

Let the findings determine the work

Better reporting should narrow down the next action. It should not leave a business with a longer list of numbers to watch.

Some findings may point towards technical access. Google’s John Mueller discussed technical SEO’s continuing relevance in an interview with SEO practitioner David Iwanow, covered by Search Engine Journal on 2 November. Google’s official guidance states that a page must be indexed and eligible to appear in Search with a snippet to qualify as a supporting link in AI Overviews or AI Mode. Eligibility does not guarantee inclusion. [5][6]

Other findings may suggest that the target is too broad. SEO platform BrightEdge’s holiday-travel research measures visibility across US English prompts from August to September 2026, rather than bookings or market share. Search Engine Journal’s 31 October interpretation identifies a possible opening for smaller publishers around inspiration questions and content drawing on human experience. That is a potential opportunity to investigate, not a guarantee for every smaller business. [7][8]

For me, the connection is straightforward: decide which audience needs the business can credibly answer, then use reporting to investigate whether its content is accessible, visible and useful to the people arriving.

A listicle might be the right format for a helpful comparison. It is a weak foundation for a strategy if the only reason for choosing it is that it currently appears to attract AI mentions.

Turn the report into a practical plan

If your AI report is producing more questions than answers, Go Up can help you work through them.

We bring search strategy, technical SEO, content, digital PR and user experience together to examine how your business is found and what happens when people arrive. That could mean checking an apparent visibility drop, improving a landing-page journey or identifying a more specific audience opportunity.

The aim is to give you clear priorities connected to your commercial goals, with an honest account of what the evidence can support.

Let’s turn your AI visibility data into a plan for what to do next.

Explore Go Up’s AI SEO services.

Sources

1. Similarweb: Being Recommended in AI Makes Users 2.5x More Likely to Visit Your Site, 23 June 2026. Downstream-visit research discussed in the transcript.

2. Kevin Indig: Does AI drive more direct traffic?, 19 October 2026.

3. Search Engine Journal: How Rank Trackers Say They Handle Suspect Google Results, 23 October 2026.

4. SEOmonitor: Google’s changes to how it serves search results and impact on SEOmonitor, provider account checked on 2 November 2026.

5. Google Search Central: AI features and your website, official guidance checked on 2 November 2026.

6. Search Engine Journal: Google Says Technical SEO Remains Key For AI SEO, 2 November 2026. Interview coverage; substantive technical guidance cross-checked against Google’s documentation.

7. BrightEdge: Holiday Travel in AI Search, August–September 2026 data; October publication.

8. Search Engine Journal: Big Brands Dominate AI Search But Leave Openings For Smaller Sites, 31 October 2026. Interpretive coverage of the BrightEdge findings.

9. Similarweb: The 2026 Generative AI Brand Visibility Index, supplied background source. Public landing page checked; distinct from source 1. No gated statistics attributed to it.

10. Go Up: Contributor roles and AI SEO services, checked on 2 November 2026.

11. Kevin Indig: Professional bio, checked on 2 November 2026.