Adapted from Duane Forrester's essay on Duane Forrester Decodes. Aurora extends his framing for marketing teams running AI-search visibility programs.
The tools are deployed. The licenses are paid. And if you are running an AI-search visibility program in 2026, you are almost certainly using AI every day — for drafts, summaries, first passes at analysis that used to take twice as long. That is real productivity. It is also not the return the investment is capable of producing, and the gap is not a tool problem. It is a mode problem.
A peer-reviewed study by Tim Gorichanaz at Drexel, presented at the 2025 ASIS&T Annual Meeting, mapped 205 real-world ChatGPT use cases into six modes: Writing, Identifying, Deciding, Ideating, Talking, and Critiquing. Writing alone accounted for 47%. Identifying — explain, summarize, look up — accounted for another 10%. Two modes dominate. Four are sitting on the table. The four left behind are the ones that decide whether AI makes you more strategically valuable or just faster at execution-layer work.
McKinsey's 2025 State of AI confirms the same pattern at the enterprise level. 88% of organizations use AI. Only 6% qualify as high performers generating meaningful enterprise-wide impact. The high performers are 3.6× more likely to have reworked their workflows rather than dropped tools into existing ones. Faster output from an unreconstructed workflow is not the same thing as better decisions from a restructured one.
The four modes most GEO practitioners are skipping
Deciding (21% of Gorichanaz's sample). Which prompts in our campaign actually have AI exposure worth prioritizing? Is the visibility gap a retrieval problem or an entity problem? Where do I spend the next sprint of budget? Most senior practitioners answer these from intuition. Used deliberately, AI in Deciding mode is a structured pressure-test of the assumptions underneath the call — applied before the call hardens. It requires more than a good question. It requires handing the model the competitive landscape, the current visibility posture, the historical performance, the strategic constraint, and then treating what comes back as genuine input.
Ideating (9%). What angles of topical authority have we failed to claim? Which third-party signals are shaping how LLMs represent our category and what would it take to shift them? What framings of our brand exist in training data that we have never addressed? These are real Ideating-mode questions. They are not "give me five blog post ideas" prompts. A real Ideating session takes twenty minutes, requires a different posture toward the tool, and produces something that cannot be replicated by anyone who did not do it.
Critiquing (6%). This is the mode with the most direct application to AI visibility work and the most organizational resistance, because it asks AI to find problems in work the team has already invested in. The weak entity claim in a strategy that sounds authoritative but is not backed by the kind of sourcing LLMs trust. The gap between what owned content says and what a well-prompted model surfaces when asked the category-level question your brand should own. The assumed premise in a GEO recommendation that made sense six months ago and is now contradicted by how retrieval patterns have shifted. Critiquing is how a senior practitioner catches what internal review missed.
Talking (8%). Rehearsal for the conversations that actually decide budget. The client call where you have to hold two causal explanations open without collapsing them into one tidy narrative. The leadership briefing where you have to explain why traditional SEO and GEO are different disciplines and need separate budgets. The agency review where you have to push back without losing the relationship. Talking mode does not produce an artifact. It produces a better practitioner walking into the room.
Where Aurora connects
The failure mode Forrester describes — execution-layer fluency masking judgment-layer hollowness — is exactly what happens when AI visibility tools become a dashboard you glance at instead of a system you reason with. Aurora is structured to push the work toward judgment:
- Citation and visibility data per prompt, per engine, per competitor is not the deliverable. It is the input. The deliverable is the diagnosis you write on top of it — which is a Deciding-mode question.
- Source-intelligence rollups of which domains the engines cite in your category are an Ideating prompt waiting to be used. Which authority surfaces could you realistically appear on, and which would shift the model's default answer if you did?
- Adaptive querying flags prompts whose results moved week-over-week. Each of those is a Critiquing prompt: what changed in our content, what changed in theirs, what changed in the model?
- Per-engine reports give you the evidence you need for the Talking-mode rehearsal — the screenshot, the cited snippet, the competitor name — before the call starts.
The compounding gap
Gorichanaz's six-mode taxonomy maps almost exactly onto the split between execution-layer work and judgment-layer work. A senior practitioner who uses AI only for Writing and Identifying is positioning themselves as an execution-layer worker at exactly the moment AI is most aggressively compressing that layer. That is not a job-displacement prediction. It is an observation about professional differentiation.
The practitioners building durable value in AI search right now are the ones using AI to make their judgment sharper, not just their output faster. The dashboards exist to support that work. They are not the work.
