Map AI Prompts to Business Outcomes

The Buyer Used to Type "Best Mattress Brantford." Now They Type Six Sentences.

An owner-operator at a Brantford mattress retailer used to know what the buying journey looked like. The prospect typed three words into Google. They saw ten blue links. They clicked, browsed, called the store. The keyword was the unit of intent and the link was the unit of conversion.

That prospect now opens ChatGPT and types something like this:

"I have a king bed, my partner sleeps hot and has lower back pain, our budget is around twenty-five hundred Canadian, we need it delivered in Brantford or Hamilton, and we hate the idea of buying online without trying it. What should we be looking at?"

Sixty-seven words. Five distinct constraints. Two locations. A direct ask for a recommendation rather than a list of links. Every keyword tool in the agency stack would miss this query because no part of it is a "keyword" in the traditional sense. The prospect is having a conversation with the engine, and the engine is being asked to act, not to look something up.

This is the entire reason Vector 3 exists. The work the diagnostic in Vector 1 surfaced and the entity-anchor work in Vector 2 stabilized now has to be pointed at something. The pointing target is the prompt inventory, the actual conversational queries real prospects send AI engines, mapped to the actual business outcomes the brand sells. Without that mapping, every later vector is optimizing into a vacuum.

Why Keyword Research Missed the Shift

Classic keyword research is built on three assumptions. First, that demand can be quantified by counting search-volume on a finite phrase list. Second, that semantically related phrases can be grouped into clusters. Third, that the optimization unit is the phrase itself.

All three assumptions weaken in AI search. Demand becomes generative; the same underlying need can be expressed in countless prompt variants, and tracking volume on each variant is both impossible and unhelpful. Clustering by semantic similarity loses information because the constraints inside a long conversational prompt (location, budget, timeline, persona, prior experience) carry more signal than the surface phrasing. And the optimization unit is no longer the phrase, it is the answer the engine needs to produce, which is upstream of any phrase list.

Search Engine Land has tracked this shift directly: the average query length on AI-mediated search has roughly doubled since ChatGPT launched, with seven- and eight-word prompts now commonplace where three-word keywords used to dominate. The longer prompts are not noise; they are the new floor of buying-intent expression. As Matt Griffin notes in client briefings, "Keyword research is still useful, but if it is the only research a brand is doing, the brand is being told what people typed in 2019. Prompt research tells the brand what people are typing right now."

The Three Layers of Prompt Intent

The Tryprofound study of fifty million-plus ChatGPT prompts revealed an intent distribution that almost no agency content has caught up to. The shorthand:

The implication for content strategy is uncomfortable. Brands that built libraries of "what is" and "how does" articles for traditional search are competing against an engine that can produce its own answers without citing the brand at all. Generative intent prompts are different; they ask the engine to build something, and the engine pulls from sources that helped it build well. A comparison page that lays out three real options against a defined set of constraints feeds the engine's generative output. A glossary entry typically does not.

The honest version of Vector 3 categorizes the brand's prompt inventory across all three layers, not just the informational one. For a Brantford mattress retailer, the informational prompts ("what is a hybrid mattress") are still worth answering; the generative prompts ("compare three king mattresses for a hot sleeper with back pain under $3,000 in Ontario") are where the citation race actually runs.

Mapping Prompts to the Buying Funnel

The prompt inventory is most useful when it is mapped to decision stage rather than to a flat keyword list. The mapping pattern that survives across most service businesses:

Each stage is its own content strategy with its own prompt inventory, its own intent class, and its own engine-citation pattern. A library that covers all five stages outperforms one that covers only the high-volume middle (solution and vendor stages) because AI engines weight context-of-decision more heavily than classic Google did.

If your content library has not been audited against an AI prompt inventory, that is the work that has to land before any new piece is commissioned. A Vector 3 prompt inventory takes about two weeks and produces the content map every later vector consumes.

The Token-Level Mechanic Behind the Shift

The deeper reason conversational prompts surface different sources than keywords do is technical, and worth understanding because it shapes content design. AI search engines and the retrieval-augmented generation systems behind them use dense passage retrieval, a method where both queries and documents are encoded as high-dimensional vectors and matched on semantic similarity rather than keyword overlap.

This is not optional reading for a content team building for AI search. The mechanic explains why thin glossary pages lose to in-depth comparison pages, why a single paragraph that addresses five constraints outperforms five separate pages each addressing one constraint, and why the FAQ schema on a page is not as important as whether the prose actually answers the questions in language that aligns with how prospects ask them. Vector 3 produces the prompt inventory, Vector 4 (Embed) is where the prose gets written to the inventory, but the mechanic that connects the two is dense retrieval.

Building a Prompt Inventory for Your Business

The deliverable that makes Vector 3 operational is the prompt inventory itself. The inventory is a structured document, typically a sheet, with columns for the prompt, the intent class, the decision stage, the priority engine (where the prompt mostly fires), the existing brand visibility on the prompt (carried over from Vector 1), and the content asset that will or does answer it.

Three sources feed the inventory. The first is the Vector 1 audit data; the category and scenario prompts that already revealed where the brand surfaces and where it does not are the seed list. The second is sales call transcripts and customer service tickets, where the conversational vocabulary the brand's actual prospects use lives in the recorded data and almost always differs from the language the marketing team writes in. The third is direct AI engine inquiry; ChatGPT and Perplexity will themselves suggest related questions and follow-up prompts when seeded with a category query, and those suggestions are a useful expansion mechanism.

The trap to avoid is letting the inventory grow to an unusable size. Three hundred prompts mapped to five real content assets is a project plan; three thousand prompts mapped to nothing is a spreadsheet that will never produce work. Keep the inventory focused on the prompts that connect to revenue, prune aggressively, and let it grow only when a new decision stage or buyer persona reveals genuinely new conversational territory.

Translating Prompts Into Content That Gets Cited

The inventory is not the work, it is the brief for the work. Each prompt in the inventory becomes a content design instruction: which constraints have to be addressed, which decision stage the reader is in, which intent class governs the answer format, which engine-citation pattern the content has to feed.

Two failure modes are worth naming because most agency content collapses into one of them. The first is keyword-density-style writing, where the conversational prompt is reverse-engineered into a phrase list and the content is written to repeat the phrases. AI engines read at the token level, not the keyword level, and the resulting content reads like 2014 SEO and underperforms accordingly. The second is the opposite, narrative-only writing that addresses none of the constraints in the prompt and expects the engine to credit the brand for tone. AI engines reward constraint-coverage, not voice. Vector 3 is the work that prevents both failures by making the constraint set explicit before the content is written.

The Mattress Miracle prompt inventory, refined over the early Formative Digital engagement, currently runs to roughly 240 mapped prompts across the five decision stages. Each prompt is tied to a specific content asset on the domain, with an explicit constraint list and an intent class. The inventory is the artifact that drives the content roadmap quarter to quarter; without it, the agency would be writing into a void. Results depend on industry, competition, and existing digital presence; the methodology of the inventory itself is what is repeatable.

The Resonate-to-Embed Handoff

Vector 3 is where the methodology pivots from diagnostic and structural work to content production. The output, the prompt inventory, becomes the input to Vector 4: Embed, where the actual prose that answers each prompt gets written. Vector 5 (Cite) then layers the authoritative citations onto that prose. Vector 6 (Structure) wraps it in schema. Vector 9 (Cluster) builds topical depth around prompt families. Each subsequent vector consumes the inventory differently, but the inventory is the shared input, the document the agency and the client both work against.

The reason this matters operationally is that without Vector 3, the later vectors fragment. Embed without a prompt inventory writes to whatever the writer thinks is interesting. Cite without a prompt inventory pulls authoritative sources for topics no prospect is actually asking about. Structure without a prompt inventory wraps schema around content that misses the constraints. The inventory is what makes the rest of the methodology coherent rather than a list of disconnected craft activities.

Frequently Asked Questions