Creators are building GEO strategies because brands now shortlist them in AI chatbots

Digiday reported on 17 August 2026 that creators are adding AI search visibility to their discoverability work, because brands and agencies have started using AI chatbots to work out which creators get cited most before striking partnership deals. Olive oil brand Kosterina told Digiday it often starts its search for potential creator partners in Claude, and agency executives told the publication that AI visibility is beginning to show up in briefs.

The report, written by Kimeko McCoy, sits in Digiday's zero-click coverage and you can read the original Digiday story for the full set of creator interviews. The piece is about creator discovery in AI search, but the mechanism it describes is not creator-specific. When the buyer of media builds a shortlist by asking a model who the credible names in a category are, generative engine optimisation stops being a consumer-facing exercise and becomes a business-to-business discovery problem.

What Digiday actually reported

Three things in the Digiday story are reported fact rather than prediction. First, Kosterina said it often starts partner searches in Claude. Second, agency executives told Digiday that AI visibility work is starting to appear in creator briefs. Third, two named creators have already changed how they publish in response, and one of them has run an audit to find out where the citations actually come from.

Digiday also cited a SponsorCX survey finding that nearly one in five consumers have already replaced traditional research with AI. That figure is the consumer half of the story and it is the half most marketers have already heard. The buyer half is newer and, for a service business, more immediately actionable.

What Courtney Johnson built

Courtney Johnson, an Austin-based creator working in the career and personal brand space, has run a website dedicated entirely to answer engine optimisation since October 2025. Digiday reported that the site repurposes her content from across social and the web as text articles built for bot consumption, carries structured data on every page, and includes a plain-text index built for LLMs. Her stated goal, in her own words to Digiday, is to help brands understand that she is an established author with credibility in the career space.

The reason the format matters is mechanical. Short-form video, carousels and in-feed posts are the bulk of a creator's output and almost none of it exists as crawlable text on a domain the creator controls. A model answering "who writes credibly about career development" is working from documents it can read. Johnson's site is an attempt to convert an existing body of work into that readable form rather than to produce anything new.

According to her own Cloudflare reporting, verified crawlers from OpenAI, Anthropic, Amazon, Apple, Huawei and DuckDuckGo all read the site. Here are the figures Digiday published, with what each one actually measures.

What those numbers do and do not show

Read the impressions figures precisely. Google Search Console impressions count times a page appeared in Google results. They are not a count of AI citations, and Digiday did not report any AI citation count for Johnson's site. The 3,048 impressions in July 2026 against 2,568 in June is growth of roughly 19 percent month on month on a small base, and 435 clicks in total since October 2025 is a modest number for ten months of publishing.

What the figures do demonstrate is that the site exists, that it is indexed, and that AI crawlers reach it. What they do not demonstrate is that answer engine optimisation produced the traffic, that the traffic produced citations, or that any of it produced a brand deal. Digiday reported the numbers as Johnson's internal reporting, and internal reporting is what it is. Anyone quoting these figures as proof that AEO works is quoting them for something they cannot support.

Colin Rocker's citation audit is the practical lesson

Colin Rocker, a New York-based full-time creator covering career development and personal finance, ran an AI citation audit with Claude in August 2026 to find out where his content was being mentioned in AI-generated answers. Digiday reported that he found he was more likely to be cited through his LinkedIn profile or his podcast appearances than through his own channels. His previous brand deals with Apple, Adobe and Meta had come through word of mouth.

That finding is the most transferable thing in the article, and it cuts against the instinct the rest of the story encourages. The reflex response to AI visibility is to publish more on your own site. Rocker's audit says the citations were living on third-party surfaces he does not own. In response he has considered pitching earned media, creating a Wikipedia page and transcribing viral video scripts. Note the framing: Digiday reported these as things he is considering, not as things he has done or measured. Do not repeat them as a proven playbook.

Why this transfers to any service business

A creator being shortlisted by an agency and an agency being shortlisted by a client are the same retrieval problem. In both cases someone asks a model an open category question, the model answers from documents it can read, and the names in that answer get the meeting. The work splits into two parts, and most companies only attempt the first.

1. Make your own material readable. Structured data on every page, and text versions of anything currently locked inside video, slides or a PDF. Johnson's plain-text index is one implementation of that idea, not a standard anyone has published.
2. Find out where you are actually cited today, then work on those surfaces. Run the audit before you commission the content, because the audit is what tells you whether the content is the constraint at all.

The audit itself is cheap. Ask a model your own brand name and ask it the category question a buyer would ask, such as who the credible agencies for a given service in a given city are. Read the sources it cites. Then check three things: whether you appear at all, which surface carries you when you do appear, and whether the pages the model leans on are ones you can influence. Directories, association pages, trade press coverage, forum threads and job or profile pages will often outrank your own site as the carrier. Those pages are then the work, and that work sits closer to public relations and influencer and partner marketing than to on-page optimisation.

Run the same audit again in a month before drawing any conclusion. A single model response is one sample, answers move between sessions, and one query is not a measurement. This is the same discipline any generative engine optimisation programme needs, and the same reason a one-off check is not worth reporting to a client as a result.

What the source did not establish

Digiday did not report any measured relationship between AI citations and partnership deals won. No creator in the story attributed a signed deal to AI visibility, and Rocker's deals with Apple, Adobe and Meta came through word of mouth. The article gave no methodology for the SponsorCX survey, no sample size and no fielding date, so the one-in-five figure should be quoted with its source attached and nothing built on top of it. It also gave no indication of how many brands or agencies actually shortlist this way, beyond one named brand and unnamed agency executives.

The story reported no data on which models cite creators, at what rate, or with what stability over time. It reported no cost for the work Johnson or Rocker has done. And it reported nothing about markets outside the United States.

What this means for Thai marketers

To be clear about the boundary: Digiday's reporting is US-based and says nothing about Thailand or the wider APAC region. Everything in this section is our reading rather than something the source reported.

Influencer selection in Thailand today runs mostly through agencies, managed talent networks and platform tools rather than through a chatbot shortlist, so the immediate buying behaviour Digiday describes is not the Thai default. The part that does apply is the supply side. Thai-language coverage of most commercial categories is thin, which means the pool of documents a model can draw on when answering a Thai-language category question is small, and a single well-structured Thai page can carry disproportionate weight in that pool. That asymmetry favours whoever publishes first, and it closes as more brands fill the gap.

The practical version for a Thai business is to run the audit in Thai as well as English, because the answer often differs by language and the sources cited almost always do. If your Thai-language presence is entirely inside video, LINE broadcasts or image posts, a model has close to nothing to read about you in Thai. Converting some of that into text on your own domain is cheaper now than it will be once the category is crowded, and it sits alongside the ordinary work of AI-aware SEO rather than replacing it.

Frequently asked questions

Are brands really picking influencers with chatbots now?

One named brand, Kosterina, told Digiday it often starts its search for creator partners in Claude, and unnamed agency executives said AI visibility is starting to appear in briefs. Digiday did not report how widespread the practice is, and it gave no figure for how many brands shortlist this way, so treat it as an early signal rather than a norm.

Is any of this happening in Thailand?

The source did not say, because the reporting is US-based and mentions no Thai or APAC data at all. Our own reading is that Thai influencer selection still runs through agencies and platforms rather than model queries, and that Thai-language citation coverage is thinner, which changes the timing of the work rather than its logic.

Do those 3,048 impressions prove that answer engine optimisation works?

No, and it is worth being blunt about it. That figure is Google Search Console impressions for Johnson's site in July 2026, which counts appearances in Google results and not AI citations, and the month-on-month rise from 2,568 is modest growth on a small base. Digiday reported no citation count and no revenue attribution against it.

Do I need to build a separate site like Courtney Johnson's?

Probably not, and the story does not claim you do. Johnson's site is one creator's implementation, built to solve a specific problem of having a decade of work locked inside social platforms. A business that already runs a content site usually has the domain it needs and is missing readable text versions of its non-text material plus structured data, not a second website.

What is the smallest useful thing I can do this week?

Run an AI citation audit for your own brand and for the category question your buyers would ask, in both Thai and English, and write down which surfaces carry your mentions. That single exercise tells you whether your problem is on your own site or off it, which is the question Rocker's audit answered for him and the one that decides where the budget should go.

Where to start

If you want to know whether your business shows up when a buyer asks a model the category question, the audit takes an afternoon and the answer usually surprises people. Relevant Audience works on AI search visibility for brands in Thailand, and the first step is always finding out which surfaces are carrying your name today.