Mapping Search Intent at Scale with LLMs
## The Bottleneck of Manual Intent Mapping
Understanding search intent, the why behind a search query, is the foundation of any successful SEO strategy. If you target a transactional keyword with an informational blog post, you will not rank. Google’s algorithm is explicitly designed to serve the format and page type that matches the user’s underlying goal.
Traditionally, SEOs categorize keywords into four primary buckets:
- Informational: The user wants to learn something (e.g., “how to brew espresso”).
- Navigational: The user wants to find a specific website or page (e.g., “Breville customer login”).
- Commercial (or Commercial Investigation): The user is researching options before a purchase (e.g., “best espresso machines under $500”).
- Transactional: The user is ready to complete an action or purchase (e.g., “buy Breville Bambino Plus”).
While identifying intent is straightforward for a dozen keywords, it becomes a massive bottleneck when you export a list of 5,000 keywords from your SEO tool. Manually tagging these in Excel or Google Sheets is an agonizing, low-value use of your time.
Large Language Models (LLMs) like ChatGPT and Claude excel at semantic categorization. By training an LLM with clear definitions and examples, you can map search intent across thousands of keywords in minutes, formatted perfectly for your spreadsheets.
Anatomy of an Intent-Mapping Prompt
To get accurate, spreadsheet-ready results at scale, you cannot simply ask an AI to “tell me the intent of these keywords.” Left to its own devices, an LLM will output inconsistent formats, invent new intent categories, or provide unprompted explanations that break your data structure.
Reliable intent mapping requires a highly structured prompt using a technique called Few-Shot Prompting. This involves providing the AI with role context, strict definitions, and a few examples of the exact input and output you expect.
1. The Role and Rules
Start by defining the AI’s persona and the strict constraints it must follow. You want it to act as an SEO expert and output only the requested format.
2. The Definitions
Do not assume the LLM shares your exact definitions of the four intents. Provide a one-sentence definition for each so the model has a clear framework to evaluate against.
3. Few-Shot Examples
This is the most critical step. By showing the LLM exactly how to handle tricky or ambiguous keywords, you align its logic with your SEO strategy.
4. Output Formatting
Instruct the LLM to output the data in a format that is easy to copy and paste, such as a markdown table or CSV format.
The Master Intent Prompt
Here is a production-ready prompt template you can use to process keyword lists.
You are an expert SEO strategist. Your task is to analyze a list of keywords and categorize each one into exactly one of the following four Search Intents: 1. Informational: The user is looking for an answer, guide, or general knowledge.
2. Navigational: The user is looking for a specific brand, website, or physical location.
3. Commercial: The user is researching options, comparing products, or looking for reviews before making a purchase decision.
4. Transactional: The user is ready to buy a product, sign up for a service, or complete a specific conversion action.
RULES:
• You must categorize EVERY keyword provided.
• You may only use the exact words: Informational, Navigational, Commercial, or Transactional.
• Do not provide any explanations or additional text.
• Output the results as a CSV format with two columns: Keyword, Intent.
EXAMPLES:
Keyword: "what is a flat white"
Intent: Informational Keyword: "blue bottle coffee locations"
Intent: Navigational Keyword: "top rated burr grinders 2024"
Intent: Commercial Keyword: "order espresso beans online"
Intent: Transactional DATA TO PROCESS:
[PASTE YOUR KEYWORD LIST HERE]Handling Bulk Data and Edge Cases
When applying this at scale, you will encounter a few practical challenges.
Context Window Limits
Do not paste 10,000 keywords into a prompt at once. Even if the LLM accepts the input, its accuracy degrades in the middle of long prompts (a phenomenon known as “lost in the middle”). For the best accuracy, chunk your data. Paste your keywords in batches of 300 to 500. Once the model outputs the CSV text, copy it, paste it into your spreadsheet, and feed the model the next batch.
The “Live SERP” Limitation
LLMs determine intent based on lexical semantics, the meaning of the words themselves. However, actual search intent is dictated by what Google’s algorithm rewards on the live Search Engine Results Page (SERP).
For example, an LLM might reasonably categorize “CRM software” as Informational (expecting a guide on “What is CRM software”). But if you look at the live Google SERP, it is heavily Commercial, dominated by listicles of “The 10 Best CRM Softwares.”
Pro Tip: Use the LLM to do 90% of the heavy lifting. Once the data is in your spreadsheet, filter by high-search-volume keywords and manually verify their intent by checking the live SERPs. The AI saves you from analyzing the long-tail keywords, freeing you up to manually verify the high-stakes targets.
Fractured or Mixed Intent
Some queries have mixed intent. A user searching “Nike running shoes” might want to browse options (Commercial) or buy immediately (Transactional). Google often serves a mixed SERP (category pages alongside reviews) for these. If your SEO strategy requires tracking mixed intents, you can modify the prompt’s rules to allow a secondary intent column (e.g., Primary Intent: Commercial, Secondary Intent: Transactional).
Apply It: Practice Mapping at Scale
To solidify this concept, you will now execute an intent mapping workflow.
The Scenario: You are the in-house marketer for a specialty outdoor gear retailer. You just exported a raw list of keywords related to “camping tents” and need to map their intent so you can assign them to either the content team (for blog posts) or the ecommerce team (for product/category pages).
Your Task:
- Open your preferred LLM (ChatGPT, Claude, etc.).
- Copy the “Master Intent Prompt” provided earlier in this lesson.
- Paste the following raw list of 15 keywords into the
[PASTE YOUR KEYWORD LIST HERE]section of the prompt. - Run the prompt and review the output.
- Copy the output and paste it into a blank spreadsheet to verify it formats correctly into columns.
Keyword List to Process:
- lightweight backpacking tent 2 person
- how to waterproof a canvas tent
- REI half dome return policy
- best family tents for heavy rain
- buy big agnes copper spur hv ul2
- tent pole repair kit
- what is the difference between a 3 season and 4 season tent
- marmot tents
- cheap camping tents near me
- how to fold a pop up tent
- top rated rooftop tents for tacoma
- coleman customer service number
- order replacement tent stakes
- glamping tent setup ideas
- outwell air tents vs pole tents
Self-Evaluation:
Look closely at how the AI categorized “marmot tents” and “cheap camping tents near me.” Did it classify “marmot tents” as Navigational (looking for the Marmot brand page) or Commercial (looking to browse)? Both are defensible, but this highlights where your human editorial judgment comes into play after the AI does the bulk sorting. Ensure your output successfully pasted into two clean columns in your spreadsheet without requiring manual formatting.