<!-- The markdown copy of https://oppalerts.com/docs/app/content/keyword_research . All pages: https://oppalerts.com/docs/llms.txt -->

[Documentation](https://oppalerts.com/docs/) / [App](https://oppalerts.com/docs/app/) / Keyword Research

# Keyword Research

`App https://oppalerts.com/app/content/keyword_research`

## What you use it for

**The words your market uses**, from two sources: model suggestions for what people search and for wording in your content.

- **Check the campaign is aimed correctly.** This is the fastest sanity check in the application: if these words do not describe your market, nothing else in the campaign does either.
- **Get the vocabulary for a page** before you write it.
- **Brief a writer** with real market language instead of your own guesses.

## How it works

![The tab strip.](https://oppalerts.com/img/docs/app/content/keyword_research/the-tab-strip.png)

*The tab strip.*

**Two tabs, both from the AI models.** One suggests terms people might search for. The other suggests phrases for your content. The lists stay separate so you can see what each suggestion is for.

### How the two model tabs are scored

Each time a keyword is named it earns **the weight of the model that named it, divided by its position in that answer raised to the power 1.5**. Position 4 divides by 8. Named first by a heavy model beats named tenth by three light ones.

The score is calculated each time the page loads. It is never stored.

### Merging

Everything merges on normalized text: lower case, punctuation stripped, spacing collapsed. So the same phrase seen ten times is one row with a count of ten, and **Seen** counts every occurrence while **Score** weighs where it appeared.

## Tabs

The page has 2 tabs.

### Tab 1: LLM Search Keywords

### What it shows

What the AI models said **people SEARCH FOR** in your market, scored and merged.

### Where the data comes from

The suggestions come from a single collection step that runs for every model.

**Check that the keywords fit your market.** Read this list before you use the reports, and check whether the terms describe what you sell and the customers you're trying to reach.

### How to use it

1. **Read the top 20 and ask whether that is your market.** That is the check.
2. **Read the Models column** when you want to know whether a keyword is a consensus or one model's idea. It counts how many different models named it.
3. **Use these as the seed for anything you buy elsewhere.**

### Tab 2: LLM Content Phrases

### What it shows

What the AI models said **the market's WRITING sounds like**: the phrases that appear in pages about this subject.

### Where the data comes from

Its own collection step, run for every model, scored the same way as the keywords tab.

**These are not search engine queries.** They are writing vocabulary, not things people type into a search box. OppAlerts never sends them to a search engine.

They are still collected, and still worth reading, because they are the language of the market rather than the language of a search box.

### How to use it

1. **Use it when writing, not when planning.** This is how to sound like the market.
2. **Give it to a writer who does not know the industry.** It is the fastest way to make a draft read as though it came from inside.
3. **Check your own page against it.** Content Optimizer does this properly, against a taxonomy.
4. **Do not build pages around these.** A phrase nobody searches will not bring anybody.

## On your own campaign

### The order to work in

1. **Open LLM Search Keywords and read the top 20.** If it is not your market, stop and fix the campaign description. Nothing else is worth reading until it is right.
2. **Open LLM Content Phrases when you brief the writing.**
3. **Take the shapes to Content Types.**

### Where this feeds

- **The campaign's reference text** is written from these lists and the other collection steps. The Opportunity Type pages and the Content Types pages search by it.

So this page is not a report at the end of the process. It is the input to most of the others, which is why it is worth reading first.

### When it looks wrong

- **Generic words, no market.** The campaign description is too vague. Rewrite it and start a new campaign.
- **The right words for the wrong buyer.** The campaign description does not say who buys. Rewrite it and start a new campaign.

## MCP

These MCP tools give the same data. Each page has every parameter and an example.

- [`llm_answers_search_keywords`](https://oppalerts.com/docs/mcp/llm_answers_search_keywords): The search keywords the AI models say people in this market search for, scored across every model and run.
- [`llm_answers_content_phrases`](https://oppalerts.com/docs/mcp/llm_answers_content_phrases): The wording the AI models use when they write about this market: the content phrases, scored across every model and run.

## API

These API routes give the same data.

- [`POST /v1/llm_answers/search_keywords`](https://oppalerts.com/docs/api/llm_answers_search_keywords)
- [`POST /v1/llm_answers/content_phrases`](https://oppalerts.com/docs/api/llm_answers_content_phrases)

## Related pages

- [Content Types](https://oppalerts.com/docs/app/content/types)
- [Content Optimizer](https://oppalerts.com/docs/app/content/optimizer)
- [All App Pages](https://oppalerts.com/docs/app/)
