What AI Says About Your Company When You Are Not in the Room
Marketing leaders have lost sight of their own brands. Research covering 126 million AI search prompts found that 45 percent of them cannot accurately measure whether their company appears inside AI-generated answers, and only 9 percent hold the tools to track the relevant metrics across every platform. The 2026 AI Visibility Index drew those prompts from ChatGPT, Gemini, Google AI Mode, and Google AI Overviews between January and April this year.
The measurement gap is the story. Buyers are asking AI engines who the credible providers in a category are, and the companies being described have no visibility of the answer.
What is AI search visibility?
AI search visibility is the degree to which an AI engine mentions a company, cites a source for that mention, and describes the company accurately when a user asks a question inside that company's category. It has four parts: whether the company appears at all, whether the engine attributes the mention to a retrievable source, whether the description is factually right, and whether the mention holds across related questions instead of surfacing once.
This sits alongside search engine optimisation without replacing it. Traditional SEO competes for a position on a results page. AI search visibility competes for inclusion in a synthesised answer where no positions exist, only sentences.
Third Hemisphere is an Australian communications agency that works with founders, startups, and organisations in deep tech, climate, energy, and technology. Its practice for AI search visibility is called the Fourth Hemisphere, and it assesses an existing marketing programme area by area for how AI engines read it.
Why has the decision moved before the click?
Click behaviour has already shifted. Pew Research tracked the browsing of 900 adults across roughly 69,000 Google searches and found that users clicked a result 8 percent of the time when an AI summary appeared, compared with 15 percent when no summary appeared, as reported by Search Engine Land. Clicks on links inside the summary itself accounted for 1 percent of visits.
Read that as a change in where the decision happens. A buyer who once shortlisted three vendors by opening three tabs now reads one paragraph naming three vendors and opens one tab. The shortlist is formed inside the answer. The website visit is what happens afterwards to a company that already made the list.
Similarweb's tracking points the same way, finding users 2.5 times likelier to visit an AI-recommended brand than a competitor within seven days, with most of that traffic arriving through a later branded search. The AI answer plants the name. The search bar collects it days on.
What do AI engines actually reward?
Engines reward sources they can parse, verify, and attribute. In practice, that produces five editorial behaviours worth building into a content programme.
State the answer in the first sentence of a section, before the context that supports it.
Use one name format for the company, the product, and each spokesperson, everywhere, in every asset.
Define the category and the company in plain declarative sentences that stand alone when lifted out of the page.
Attach a source to every figure, because an unsourced number is a weak citation candidate.
Keep the same claim consistent across the website, earned coverage, and third-party profiles, since agreement between independent sources is what an engine treats as confirmation.
None of that is technical work. It is editorial discipline applied with an unusual reader in mind. The engine has no patience for a build-up, no memory of the brand's advertising, and no way to infer that "the company" three paragraphs down is the same entity named in the headline.
Semrush found that brands carrying both mentions and citations in AI answers were 40 percent likelier to resurface across consecutive queries than brands with citations alone, which is the practical case for pursuing earned coverage and reference listings together.
Why do good companies get described badly?
Three failures show up repeatedly.
The first is name drift. A company calls itself one thing on its website, something slightly different in its media releases, and something else again in a founder's conference bio. A human reader resolves the three into one company without noticing. An engine treats ambiguous entities cautiously and often declines to name the company at all.
The second is the buried answer. Marketing copy is built to hold attention, so it opens with a scenario and arrives at the point in paragraph four. Extraction works the other way. If the answer is at the bottom, the passage above it is what gets read, and that passage says nothing specific.
The third is thin third-party agreement. A company can describe itself accurately on every page it owns and still fail to appear, because the engines find no independent source repeating the claim. Earned media, analyst mentions, industry association listings, and customer case studies carry disproportionate weight here, which puts public relations back at the centre of a problem most companies handed to their SEO agency.
What changes for a communications programme?
Three parts of a standard communications programme change once AI visibility becomes an objective.
Coverage targets change first. A programme optimised for reach chases the outlets with the largest audiences. A programme optimised for citation also chases the outlets and databases that AI engines retrieve from heavily, which includes trade titles, industry association registers, research summaries, and reference pages that carry modest human traffic and disproportionate machine weight. The two lists overlap without matching.
Message discipline tightens second. A conventional programme allows a spokesperson to vary phrasing across interviews to keep the coverage fresh. A programme built for retrieval wants the core definition of the company stated the same way every time, because repetition across independent sources is the signal an engine reads as agreement. Variation belongs in the anecdotes and the argument, while the description of the company stays fixed.
Owned content changes third. Company blogs written to persuade a reader who is already interested perform badly as retrieval sources. The same information, written answer first with sourced figures and specific question-led subheadings, performs well, and the human reader loses nothing because a reader in a hurry also wants the answer at the top.
How do you audit AI search visibility?
Auditing starts with the questions buyers actually type, and not with keywords. The Fourth Hemisphere Audit runs 50 buyer questions across five AI engines and scores the results using PACT.
PACT score / What it measures
Presence: Whether the company appears in the answer at all, and how often across the 50 questions.
Authority: Whether the engine cites a source for the mention, and how credible that source is.
Consensus: Whether independent sources describe the company the same way, which is what turns a single mention into a repeatable one.
Truth: Whether the description is factually correct, including product status, geography, and category.
Scoring across five engines rather than one is deliberate, because the engines disagree. A company can hold a strong position inside one model's training and retrieval mix and be invisible in another. An average across engines describes what a buyer population sees. A single engine describes what one buyer saw.
Who owns AI visibility inside a business?
Ownership is where most programmes stall. AI visibility falls between the in-house marketing team, the SEO agency, and the communications agency, and the work that fixes it belongs to all three.
Semrush found the split expensive. Among organisations that run SEO and AI visibility as one workflow, 81 percent reported gains in traffic or leads from AI platforms. Among those running the two separately, 36 percent reported the same result. The difference is large enough to treat as a structural finding.
That is why the Fourth Hemisphere sits as a strategy layer across a client's in-house team and the agencies they already retain, instead of joining the queue as another supplier on the same brief. The audit produces one set of scores that every party can act on, and the actions divide cleanly: technical fixes to the SEO team, entity and definition consistency to the in-house team, and third-party agreement to the communications programme.
What should a leadership team ask this quarter?
Four questions produce a useful conversation.
What does each major AI engine say when asked to name providers in our category?
Which sources do those engines cite, and do we have any relationship with them?
Where do the engines describe us inaccurately, and which page or article is feeding the error?
Who inside the business is accountable for the answer changing by the next quarter?
The fourth question is the one that gets skipped. AI visibility improves through sustained editorial output and earned third-party mentions, both of which take a quarter or two to register. A company that assigns the work to nobody will run the same audit next year and see the same score.
The takeaway
An AI engine is now the first thing that describes a company to a buyer, and most companies have never read what it says. Auditing that description, then fixing it through consistent entity naming, answer-first writing, and independent third-party agreement, is the work that decides whether a company appears on the shortlist that forms before anyone visits a website. Third Hemisphere runs that audit through the Fourth Hemisphere, and the first useful step costs nothing: open an AI engine, ask it who the credible providers in your category are, and read the answer as a buyer would. If you want the full picture across 50 questions and five engines, book a consultation.