How AI represents the company

Readers may see the answer without visiting the source

Citations give readers the option to open the original documents. They do not require it. Pew Research Center’s July 2025 analysis observed fewer clicks on standard results after Google searches with an AI summary than after searches without one.

The analysis covers browsing by 900 US adults in March 2025, with search results collected afterwards in April. It does not measure a particular company’s lost traffic or establish a causal effect. It illustrates why some reading may happen before any visit to the site.

Begin with questions that matter in the market

Monitoring should begin with actual situations: understanding a technology, examining investment constraints or comparing ways to solve a problem. Asking for the industry’s leading companies may produce a flattering list while revealing little about the company’s role in those decisions.

Communications can develop these questions with experts, sales teams and account managers. The scope should include what arises before a tender and while assessing an offer: required capabilities, risks, delivery conditions and available evidence. This defines the monitoring exercise; it does not reconstruct every buyer’s behaviour.

The review then concerns associations: does the company appear in its current activity, for a capability it can substantiate, through relevant sources? Absence from a priority question and an inaccurate description call for different actions. Representation within the market matters beyond the number of mentions.

Correct the documents within your control

Inconsistent service descriptions, old biographies and undated figures can be corrected at source. Check metadata and related pages too: an updated article can coexist with an outdated summary.

Google says its AI features require no special optimisation or structured data. Technical access and established search requirements still need checking. Clear, consistent information also helps human readers, while selection remains the service’s decision.

Google’s guide, updated in July 2026, also specifies an eligibility condition: the site must be included in generative AI features through Search Console. This setting concerns inclusion, not guaranteed citation. It needs checking separately from content quality and indexing.

Read what the answer attributes to the company

Then read what is actually attributed. An answer may mention the brand while drawing on another source, or repeat a figure without its scope. The correction depends on the finding: outdated information, an unclear page, poorly documented data or a contribution that remains hard to find.

I distinguish four objects in the record: the capability associated with the business, the claim made about it, the page actually cited and the evidence available on that page. A citation may sit beside a passage without supporting every word. Return to the original document and its scope.

The remedy depends on where an error originates: update a controlled document, alert its publisher or make a missing explanation accessible. More pages targeting the same keyword do not resolve contradictory sources.

Measure citations without assuming a commercial effect

Start with a complete trace: service, question, date, answer and cited pages. A screenshot without the question can mislead. Being cited for a definition, a supplier comparison or a controversy does not tell the same story about the company’s place in the answer.

After an update, repeat a defined set of questions and keep several observations. An accurate citation for a capability that matters can guide editorial work; it does not prove preference or a buying decision. The record should connect the question, the activity, the evidence and the possible use of the answer.

This work connects the site with media relations, expert publications and other accessible sources. Information may be clear on the company’s site but missing from places its audiences consult. Improving its presence requires a contribution that is accessible and relevant, without promising what an engine will select.