Intent classification
Every prompt is sorted into one of five intent types before anything is asked, so a buying-intent citation is never averaged with a passing mention in an explainer.
A citation on "which accounting software should I buy" and a citation in "how does cloud accounting work" are not the same commercial event. Intent classification is what keeps them apart.
What you'll learn
- The five intent types, with examples
- Why the mix matters more than the total
- How to use intent when you are deciding what to fix
The five types
| Intent | What it is | Example |
|---|---|---|
| Local | Place is part of the question | Best physiotherapist in Geelong for a running injury |
| Buying | Somebody is about to choose | Which project management tool should I buy for a small team? |
| Comparison | Two or more options weighed | Xero or MYOB for a construction business? |
| Informational | Learning, not yet choosing | How does cloud accounting actually work? |
| Branded | Your own name is the subject | What do people say about [your brand]? |
Classification happens before any engine is asked, so it is a property of the prompt rather than a guess made about the answer afterwards.
Why the mix matters more than the total
A 60 percent citation rate made entirely of informational prompts is worth less than a 25 percent rate on buying and comparison prompts. The first says engines will quote you when explaining a concept. The second says engines recommend you when somebody is choosing.
So read your rate by intent before you read it in total:
- Strong on informational, weak on buying. You are seen as a source, not as an option. Usually an authority and positioning problem: engines know what you know, not what you sell.
- Strong on buying, weak on comparison. You get named when the buyer already knows the category. In head to heads somebody else's framing is winning.
- Weak on local, everything else fine. Your place signals are thin. Check Authority in Readiness, and check that your prompts actually contain the place.
- Branded prompts returning something you dislike. That is reputation, and it is the one case where the source is usually a page you do not own. Sources will tell you which one.
Using intent to choose what to fix
- Sort the losses by intent. Losing a comparison prompt costs more than losing an explainer.
- Match the fix to the type. Comparison losses are usually answered by a page that makes the comparison honestly, including where you are not the right choice. Local losses are answered by place and trust signals. Informational losses are answered by a clean, extractable answer near the top of a page.
- Check the intent mix of the prompts you are adding. A prompt set that drifts towards informational drifts away from revenue.
Try this in Outercite
Open Prompts and group by topic, then read the intent of the prompts you are losing. If they are mostly informational, your rate is less of a problem than it looks. If they are mostly buying, it is more.
Related
Your first prompts
Choosing a set with the right intent mix.
Winning local and buying-intent queries
The playbook for the two highest-value types.
Data definitions
How rates are computed within a type.
Last updated on
The six-level stack
The six levels that decide whether AI names you: four on your site, two in the loop you run. They fail bottom up, so fix them in order.
Competitors and share of voice
Share of voice is your portion of the naming across your tracked prompts. Here is how it moves, and what each kind of movement means.