Is AI search commercially valuable when we can’t reliably measure its influence?

Two October analyses sharpened this problem. Search consultant and researcher Marie Haynes examined referral data across 100 websites in an analysis published on 23 October. Organic growth advisor Kevin Indig, author of Growth Memo, separately investigated whether AI visibility correlated with subsequent branded search or direct traffic, publishing his findings on 19 October.

Meanwhile, Semrush’s 21 October analysis found AI Overviews appearing more frequently across its sample of commercial queries. The question of AI’s commercial value is increasingly difficult to separate from everyday SEO.

The challenge is to take measurable results seriously without assuming they tell the whole story. We also need to resist explaining every gap in our analytics as hidden AI influence.

Google’s traffic advantage deserves serious weight

Haynes analysed 12 months of GA4 data across 100 websites, covering more than 145 million sessions. ChatGPT sent traffic equivalent to 0.62% of Google organic’s volume. Including other large language models brought that to 0.73%.

Among the 55 sites tracking conversions, ChatGPT visitors converted at 13.8%, compared with 3.02% for Google organic. Yet Google organic generated 27 times more conversions than all the LLMs combined. Those findings describe this dataset, rather than benchmarks every business should expect. [1]

A high conversion rate is encouraging. It does not automatically make a small source of visitors your biggest commercial opportunity. The difference in volume matters when deciding where to put the next pound or hour.

Indig investigated a familiar explanation for missing AI traffic: perhaps recommendations lead people to return through branded search or direct visits. Across four brands, he could not establish a reliable relationship between AI visibility and subsequent branded search or direct traffic. A separate analysis of 12 sites found that much of the apparent growth in direct traffic was bots. [2]

His findings leave room for an effect he could not isolate. They do not prove that AI has no influence. But they challenge the assumption that a substantial hidden contribution must be there.

Referral traffic may answer only part of the question

My reservation is with using immediate website visits as the complete test of AI search’s value.

Someone could research a business through an assistant, consider its recommendation, then search Google for the business days later. Analytics might record the eventual visit without revealing the earlier research.

The comparison with a Super Bowl advert is useful here. Judging the advert solely by immediately attributable website visits would overlook the possibility that it influenced recognition or a later decision. The same measurement problem could apply to AI recommendations.

That analogy explains a possible limitation in our measurement. It does not establish that an AI mention produced a commercial return.

Understanding that audience is the next task. Interest in AI gives us a reason to investigate its role, rather than evidence that an investment will pay off.

Evidence of influence is not the same as reliable attribution

An older study provides a useful counterpoint. Published on 23 June, research from digital intelligence platform Similarweb followed users asking ChatGPT about finance, travel and beauty, then examined their behaviour over the following seven days. It excluded users who had named the brand in their prompt or visited its website in the preceding four weeks.

Similarweb reported that users were, on average, 2.5 times more likely to visit the recommended brand’s website than a competing brand’s. Much of the subsequent traffic arrived through branded search. [3]

That does not mean an AI mention will multiply any website’s total traffic by 2.5. It describes behaviour within a particular study and does not establish causation on its own.

Nor does it neatly cancel out Indig’s research. Different samples and methods can reveal different aspects of the relationship. Together, the studies illustrate the tension between observing downstream behaviour and reliably identifying its cause in a business’s analytics.

If research begins in an assistant and continues through Google, both stages deserve attention. That is why I think the reporting perspective needs to widen: the days of having your SEO report just filtered to organic are over.

The difficulty remains attribution. Seeing AI visibility and branded traffic rise together would give us something to investigate, but would not demonstrate that one caused the other.

Follow branded searches beyond the homepage

The homepage is an obvious place to examine what happens after someone has researched a business elsewhere. A visitor arriving with a particular need should be able to find the relevant next step.

Which terms accompany the brand name? Where do those searches lead? What can visitors find and do when they arrive?

Those patterns can be considered alongside evidence of AI visibility without automatically being attributed to it.

There is another complication: the same entry page can serve very different audiences. Existing customers looking for a login and prospective customers considering a purchase may both arrive through the homepage. Branded traffic needs interpretation before it can support a claim about new demand.

Examine what independent sources actually substantiate

Understanding a journey also means asking why an assistant might recommend a business for a particular need.

My working hypothesis is that relevant third-party corroboration can help explain differences in AI visibility. A business may describe itself broadly on its website, while independent sources support a more specific account of its strengths.

That raises a useful question for digital PR. An expert’s opinion on a topical story and an independent assessment of a product’s suitability do different jobs. My view is that the latter may be more directly relevant when an assistant is answering a buying question.

That qualification makes the next step more disciplined. Examine the sources cited, the claims they support and the context of the query. Relevant corroboration is a hypothesis to investigate, rather than a basis for promising that press releases generate AI citations or dismissing reactive commentary.

Commercial AI Overviews make this an SEO question too

Google’s own results make it increasingly difficult to treat AI visibility as a separate concern.

Semrush tracked a fixed panel of 400 commercial keywords across banking, software and ecommerce using US desktop data. AI Overview coverage increased from 45.75% in September 2025 to 88% in August 2026. Review queries rose from around 30% to nearly 90%, while “best X for Y” queries reached 96.12%. [4]

Those percentages describe a selected sample. They do not establish that almost every commercial search now produces an AI Overview. They do challenge the idea that AI summaries concern only informational content at the top of the funnel.

Visibility still needs a commercial interpretation. Exposure, a later search and a conversion are different observations. Connecting them requires more evidence than showing that the brand appeared in an answer.

Scrutinise the tools used to measure visibility

Even the data informing those investigations needs care.

On 22 October, SEO software provider Ahrefs reported abnormal results being served to suspected search-result data providers. It described filters to reject suspect results, requests for replacements and work to clean up historical data. This is the provider’s account, rather than independent confirmation of Google’s motive. [5]

For me, that strengthens the case for starting with Search Console when assessing actual Google search performance and checking tracker movements against it before drawing conclusions.

AI tracking presents a related question about what the measurement represents. Repeating a prompt provides a view of responses to that prompt. It does not establish how many prospective customers ask the question, what follow-up information they supply or the answers they eventually receive.

A tracked prompt can be useful without being a reliable measure of demand. Before treating movement in its visibility score as a commercial opportunity, we need to understand how closely it reflects the audience’s research.

PHP security issues put temporary downtime in focus

The PHP security issues disclosed in October brought a more immediate concern for website owners. On 21 October, the PHP project released fixes for two critical vulnerabilities that could allow attackers to execute code remotely on affected servers. Its advisory urged users to update immediately. [10]

CERT-EU advised organisations to patch affected systems immediately and, where that was not feasible, consider isolating internet-facing applications from public networks until a fix had been applied and verified. That is the context for the SEO question: how should a website signal temporary unavailability while security work is carried out? [11]

For a short closure of one or two days, Google recommends returning a 503 Service Unavailable response while keeping robots.txt accessible. Its guidance also recommends static HTML and a Retry-After header. [6]

A 503 signals temporary unavailability; it does not fix the vulnerability or guarantee that rankings will remain unchanged for two days. Prolonged unavailability can affect indexing. The priority is to secure the affected system and restore service safely, using the appropriate status code during any necessary downtime.

Turn search uncertainty into a clear plan

Knowing that search is changing is one thing. Deciding what deserves your budget, what to measure and what to do next is harder.

Go Up can help you make those decisions. We bring SEO strategy, technical expertise, content, digital PR and user experience together to assess how your business is being found, what supports its visibility and where the opportunities lie across Google and AI search.

That could mean understanding your branded search journeys, examining the sources behind AI recommendations or identifying which pages need to work harder when visitors arrive. The aim is to turn a complicated picture into clear priorities connected to your commercial goals.

If you are wondering whether AI search deserves more of your attention, or how to assess the work you are already doing, let’s talk. We can help you decide where to focus and build a search strategy around the evidence available.

Explore Go Up’s AI SEO services.

Sources

1. Marie Haynes: GEO Is A Distraction, 23 October 2026.

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

3. Similarweb: Being Recommended in AI Makes Users 2.5x More Likely to Visit Your Site, 23 June 2026.

4. Semrush: AI Overviews across commercial search modifiers, 21 October 2026.

5. Ahrefs: An update on Google’s poisoning of search results, 22 October 2026.

6. Google: Temporarily pause or disable a website.

7. Marie Haynes: Professional bio.

8. Kevin Indig: Professional bio.

9. Go Up: Contributor roles.

10. PHP: Critical security advisory, 21 October 2026.

11. CERT-EU: Critical Vulnerabilities in PHP.