The method · § 00

The LLM is not a black box.

We built a model of how foundation models choose which brands to recommend. ktau turns that understanding into the precise plan that moves them — query by query, attribute by attribute, brand by brand.

§ 01 · The premise

Measurement is the wrong end of the funnel.

The GEO category has converged on the same instrument: a dashboard. Citations, share of voice, sentiment. A picture of where you stand. Useful — but inert.

To move the LLM, you need a different kind of instrument. One that doesn't just describe the answer, but predicts what would change it. That is the work ktau does, and it is what separates strategy from monitoring.

What the category builds

A picture of the answer.
Citations, mentions, share of voice, sentiment. Tells you you're losing — leaves the rest to you.

What ktau builds

A model of the decision.
A working theory of how the LLM is forming this answer — and the specific moves that would change it.
§ 02 · The model

From the outside, one sentence. From the inside, a surface.

From the outside, an LLM's recommendation looks like a single fluent sentence. From the inside, it's the outcome of thousands of weighted signals: training data, retrieval candidates, source authority, attribute matches, conversational context. The output is fluent; the decision is probabilistic.

We treat that decision as something to be modeled, not guessed at. For every category we work in, ktau builds a brand-aware model of the recommendation surface — what the LLM is drawing on, where the leverage points sit, and what would tip the answer in your direction.

Figure 01 · the recommendation surface
A modeled recommendation surface for one category. Heat indicates where the LLM is currently most likely to mention your brand; leverage points are the moves that would shift the surface in your favor most efficiently.
§ 03 · How we probe

Three instruments, run continuously.

We don't audit once. We probe the recommendation surface every day, every model, every locale. The point is to keep the model live — because the LLMs underneath it are.

01
Query probing at scale
We simulate the buyer-intent queries that matter in your category — thousands of them, every week — and observe how foundation models actually respond. The output is a statistically grounded picture of where you appear, where you don't, and against whom. Not vibes. Distributions.
02
Source influence mapping
For every winning and losing query we trace the sources the LLM relied on. A handful of sites move recommendations in any category. Most don't. Knowing which is the difference between a content strategy and a wishlist of blog topics.
03
Attribute-level visibility modeling
Brands don't win or lose categories whole. You win on pricing and lose on enterprise fit; you dominate ease of use and disappear on integrations. We model recommendation at the attribute level, because that's where the LLM actually decides — and where the most efficient lift lives.
§ 04 · From signal to strategy

Granularity is the point.

A brand with zero AI presence has a different problem than a category leader, and the answer is not "publish more." Emerging brands need category-level content first. Established brands need attribute-level depth. Dominant brands need defense.

ktau detects which problem your brand actually has, then prescribes accordingly — what to publish, where to publish it, and what each piece is forecasted to move. This is the line between us and content factories. We don't ask you to publish more. We tell you which one piece will most efficiently move your weakest attribute — and we measure the effect.

"We don't ask you to publish more. We tell you which one piece will move the LLM most — and we measure it."

§ 05 · Two effects

One content engine. Two windows of impact.

Every piece of content produced through ktau is engineered for two effects. Most of the category optimizes for the first. We optimize for both — and the second is where durable AI visibility actually lives.

Effect · 01

Retrieval-time

When the LLM reaches out to the web mid-answer, your content is the source it picks. Citation share rises. You appear in the response. Lift is measurable in weeks.

Time-to-impact: days–weeks
Effect · 02

Training-time

When the next foundation model trains, your content is part of what it absorbs. The model begins recommending your brand without any search at all. Visibility compounds across model releases.

Time-to-impact: compounds permanently
§ 06 · The loop

Not a funnel. A tightening loop.

Decode → Direct → Deliver, but not in a straight line. Every published piece updates the model. Every model update refines the next prescription. Over weeks the loop tightens, and the LLM's answer in your category stops describing the market and starts describing your brand.

§ 07 · What the method produces

Two brands. Two starting points. One method.

Different brands have different problems. The method makes the difference between them obvious — and actionable.

Case · Emerging brand · No prior AI presence
40 days
From invisible to top-5 in category. The method started with broad category-level content, then narrowed to attributes once the foothold was secured.
Case · Multinational · Enterprise scale
2 weeks
To category dominance on targeted attributes. Here the work was surgical — high-leverage source publishing aimed at the specific features the brand was losing on.
§ 08 · Start

Run the method on your brand.

Send us your brand and category. Within 72 hours we deliver an attribute-level visibility map, the leverage points we'd attack first, and a forecast of what they would move.