Dentsu and Adobe Bet on GEO as AI Search Eats SEO

Any platform lead who has watched organic search traffic bleed into ChatGPT summaries over the last eighteen months knows the exact shape of this problem. You still rank on Google, but the answer box now lives inside an LLM, and nobody on your team can tell you whether your brand made it into the context window. On July 27, Dentsu Digital and Adobe shipped a product aimed straight at that blind spot.

What Happened

As BigGo Finance reported, Dentsu Digital announced on July 27 the launch of a Generative Engine Optimization (GEO) support service built on top of Adobe's newly branded "Adobe LLM Optimizer." The pitch is straightforward: analyze brand mentions across multiple large language models in real time, visualize competitive gaps, and provide end-to-end support from strategy design through ongoing operations.

This is not Dentsu Digital's first swing at the category. The agency has been running a "GEO Consulting Service" since May 2025, roughly a year of billable engagements with what it describes as numerous clients. The new offering wraps that consulting muscle around Adobe's measurement layer, which handles the actual telemetry: brand mention tracking across LLMs, trend visualization by prompt, and competitive gap analysis.

The service ships as a four-step process. Step one: pick which AI tools to monitor and design the prompts and keywords that matter. Step two: run cross-sectional monitoring with Adobe LLM Optimizer against those prompts. Step three: plan and execute a GEO strategy across what Dentsu calls "triple media," Earned (word-of-mouth and reputation), Shared (social), and Owned (company websites). Step four: measure score deltas before and after implementation, then loop through PDCA.

Dentsu framed the collaboration around "digital marketing solutions that fuse advanced technology with advanced operational capabilities, contributing to corporate business growth in the AI search era." Translation: consumers are querying ChatGPT and Gemini directly, and someone has to answer the CMO's question about whether the brand shows up in those answers. The rollout is aimed squarely at the Japanese market, where Dentsu expects the partnership to produce concrete case studies first.

Technical Anatomy

Strip away the marketing wrapper and this is a measurement and attribution problem, not a generation problem. Traditional SEO has a decade of tooling around a well-understood substrate: Google's crawler, an index, a ranking algorithm, and a SERP you can scrape. GEO has none of that. The "engine" is a probabilistic model whose training cutoff, retrieval augmentation, and system prompt all vary by vendor, by product tier, and by week.

So what does Adobe LLM Optimizer actually do? Based on the source facts, it performs three functions: real-time analysis of brand mentions and citations across multiple LLMs, trend visualization broken down by prompt (query), and competitive gap analysis. In practical terms, that means the tool is running structured prompts against a matrix of models, parsing the responses for brand references, and diffing the results over time. Anyone who has built LLM eval harnesses on top of the OpenAI API or Anthropic's endpoints will recognize the shape. Prompt fan-out, response parsing, longitudinal storage, dashboard.

The hard part is not the harness. The hard part is the interpretation layer. LLM outputs are non-deterministic. Run the same prompt twice at temperature above zero and you get different citations. Run it against ChatGPT with browsing enabled versus disabled and you get different brand mentions entirely. Any credible GEO measurement stack has to account for sampling variance, model version drift, and the retrieval layer sitting behind each product. Adobe's tool presumably handles some of this. The source doesn't detail the sampling methodology.

The "triple media" framing is where the consulting billable hours live. Earned, shared, and owned media influence what an LLM ingests during training and, more importantly, what a retrieval-augmented system pulls at query time. If your brand's Wikipedia entry is thin, if your owned content is JavaScript-rendered garbage that crawlers cannot parse, if your social presence is inconsistent, you will not show up in the context window. My take: the real product here is a feedback loop between measurement and content operations, and that loop is where teams either extract value or waste six figures.

Who Gets Burned

The uncomfortable read: this launch is a leading indicator that the entire SEO industrial complex is being repriced, and the teams downstream of that repricing have roughly two quarters to react. If you run a marketing engineering function at a mid-market fintech, iGaming operator, or ad-tech vendor, your dashboards almost certainly do not measure LLM citations today. That gap is now a competitor's talking point.

Japanese enterprises are the first cohort in the crosshairs, since Dentsu explicitly expects the partnership to generate concrete use cases in that market. Expect the playbook to travel. European agencies with Adobe partnerships will ship equivalents within months. Operators in regulated verticals like iGaming will find this particularly awkward because their organic acquisition funnels already run through affiliate networks and comparison sites, both of which are exactly the kind of Earned Media signal an LLM weighs heavily when answering "which sportsbook accepts crypto in Ontario."

Fintech is the more interesting exposure. When a consumer asks Gemini "what is the best neobank for freelancers in Germany," the answer determines a real acquisition. Production incidents I've seen around SEO ranking changes typically cost teams weeks of unplanned engineering work. GEO drift will be worse, because the model updates are opaque and the remediation path is not a robots.txt tweak.

The teams who get burned hardest are the ones with strong SEO ops, weak brand PR, and no structured data discipline. Their owned content ranks fine on Google and gets ignored by LLMs. Meanwhile, competitors with mediocre SEO but strong third-party mentions win the AI citation game by accident. That inversion is going to surprise a lot of CMOs in Q4.

Playbook for AI Development

Concrete actions for this week, if you own any surface that depends on discovery traffic:

First, stand up your own measurement harness before you buy anyone's product. A weekend of engineering time gets you a script that runs a fixed set of brand-relevant prompts against ChatGPT, Gemini, and Claude, logs the responses, and parses mentions. You do not need Adobe LLM Optimizer to establish a baseline. You need discipline. The Anthropic docs and OpenAI's platform docs both cover the basics of programmatic access.

Second, audit your structured data. Schema.org markup, clean HTML, server-rendered content, and a real sitemap still matter, arguably more than before, because retrieval systems reward parseable sources. If your product pages are React-rendered without SSR, that is now a citation problem, not just an SEO problem.

Third, treat third-party mentions as an engineering-adjacent concern. Wikipedia presence, industry directory listings, and podcast transcripts feed the training corpus and the retrieval index. Coordinate with PR instead of ignoring them.

Fourth, do not sign a multi-year GEO consulting contract based on a category that the source itself describes as a nascent concept. Buy quarterly. The measurement methodologies will change as model vendors change their retrieval behavior. Lock-in on a fast-moving substrate is how teams end up paying for last year's assumptions.

Fifth, name an owner. GEO sits between SEO, brand, PR, and platform engineering. Without a single accountable lead, nothing ships.

Key Takeaways

• Dentsu Digital and Adobe launched a GEO service on July 27 built on Adobe LLM Optimizer, targeting brand visibility inside LLM answers.
• The four-step process (prompt design, cross-LLM monitoring, triple-media execution, PDCA measurement) is a measurement-plus-consulting play, not a generation tool.
• Dentsu's GEO consulting has been running since May 2025, so the Adobe partnership is a productization of existing billable work.
• Teams with strong SEO but weak Earned Media are the most exposed, because LLM retrieval weights third-party citations heavily.
• Build your own baseline harness this quarter before committing budget to any vendor in a category the source itself calls nascent.