AI keyword research: how to automate finding keywords

Francesc Sánchez Francesc Sánchez — CEO and AI Consultant
Updated:
hacer keyword research con IA desde el portátil

Anyone running an online store or a services website usually does keyword research once, when the site launches, and then forgets about it. A year goes by, the catalog changes, people search differently, and the blog keeps covering the same handful of ideas as always. Artificial intelligence changes that part of the job, the part about figuring out how people search, which is exactly the most tedious part and the one that used to take hours in front of a spreadsheet.

What AI doesn’t change is the data. A language model can suggest phrasings, group searches by what the user wants and discard the ones that have nothing to do with your business, but it doesn’t know how many people search for each thing. For that you still need a source of search data. This guide explains how to combine the two: first by hand with ChatGPT or Claude, then with SEO tools that already have AI built in, and finally with an n8n or Make workflow that does it on its own every month. At the end you’ll also see when none of this is worth setting up.

Generating keyword ideas with ChatGPT or Claude

Expand each seed into variations and synonyms

A seed keyword is the starting term, the one you would use to describe what you sell. If you run a specialty coffee shop, the seed might be “coffee beans”. The problem is that your customers don’t always call it that: some search for “coffee for moka pot”, others for “buy freshly roasted coffee”, others for the name of a specific origin.

writing down the variations of each seed keyword suggested by the assistant

This is where an assistant like ChatGPT, Claude or Gemini really pays off. You give it the seed, explain what you sell, to whom and in which country, and ask it for the different ways a person would search for that on Google. It’s worth asking it to separate synonyms, variations with a brand or origin, comparison searches and searches with purchase intent, because if you don’t, everything comes out mixed together in a list that’s hard to review afterwards.

Two details make the difference in the result. The first is giving it business context: the model doesn’t know whether you sell to consumers or to restaurants and cafés, and the list changes a lot from one case to the other. The second is asking it to write the way people type into the search bar, with spelling mistakes, without accents and in short phrases, because otherwise it tends to return catalog vocabulary that nobody types.

Find the long tail and question-style searches

Long-tail searches are those of four, five or six words, more specific and with less competition, such as “which coffee to buy for a reusable capsule machine”. Few people search for each one, but those who do know what they want, and for an SME they’re often the only ones where you can compete with the big brands.

Ask the model to come up with, for each variation from the previous step, the questions a customer would have before buying: how it’s used, how long it lasts, how it differs from the alternative, whether it works for a specific case. Question-style searches are also the ones that fit best in blog posts and in the answers of AI search engines. If you also ask it to add search volume or difficulty to each keyword, it will, with numbers that look reasonable and don’t come from any database. ChatGPT and Claude have no access to Google’s search data, so those numbers are useless for making any decision.

General-purpose assistants or SEO tools with built-in AI

AI keyword research tools fall into two families, and choosing the wrong one is the most expensive mistake in the whole process, in money or in time.

General-purpose assistants, ChatGPT, Claude or Gemini, are good at generating ideas, rephrasing and classifying, and they have no data. They’re useful for the creative part of keyword research and for sorting lists that come from somewhere else. If you already pay for one of them for other tasks, you don’t need anything else for this part.

SEO tools with built-in AI do the opposite: they start from a database of searches and add AI features on top. Ahrefs’ Keywords Explorer has AI suggestions that generate new ideas, plus keyword clustering and search intent detection, all from the Standard plan up according to its pricing page in November 2026. Semrush has the Keyword Strategy Builder, which groups keywords into topics and clusters by comparing which pages rank for each search, and requires a paid subscription. Here every idea comes with its real volume, and that’s what you’re paying for.

There’s a third route that gets used less and is the one that matters if you want to automate: keyword data APIs. DataForSEO, for example, sells search volume, difficulty and intent per query, on a prepaid model where the minimum top-up is 50 dollars, according to its pricing page. Ahrefs includes API access from the Lite plan, with a lower usage limit on that plan than on the higher tiers (we checked in November 2026, and it’s one of the things that changes most). The Google Ads Keyword Planner is free, but accounts with no ad spend only see very broad ranges instead of volumes, so it’s good for finding ideas and not much use for prioritizing.

Our rule of thumb for an SME owner is simple. If you do keyword research a couple of times a year, a general-purpose assistant plus the free trial or basic plan of an SEO tool is enough. If you want the process to run on its own, you need a model and a data API, and that’s what we set up in the next section.

How to build a keyword research workflow in n8n or Make

n8n and Make are tools for chaining apps together without coding, or with very little code. Each step of the process is a node (in Make they’re called modules), and the data passes from one to the next. The workflow described here can be built in either; the node names used are n8n’s.

The starting point is a Google Sheets spreadsheet with your seed keywords, one per row, with a column for the country and another for the product or service category it belongs to. The first node of the workflow reads that sheet. That way, when you launch a new product line, adding its seed means writing one row, and you don’t have to touch the workflow.

the spreadsheet of seed keywords that feeds the keyword research workflow

Generate variations and look up volume and difficulty in an API

The second node calls the model, OpenAI or Anthropic, with the same request you’d make by hand in the chat. You pass it the seed and the business context, and ask it to return the variations, the long tail and the questions in a fixed format (a JSON list) so the next node can read them without parsing free text. The format breaks every now and then, so it’s a good idea to turn on the node’s retry option or add one that checks the response can be read.

With the list split out, an HTTP Request node sends it to the data API you’ve chosen, DataForSEO or Ahrefs, and gets back each keyword’s monthly volume and difficulty. Zero-volume keywords can be discarded here or stored separately, because for a local or niche website they’re sometimes exactly the ones that matter, and the tools don’t always record very low-demand searches. This is the step that costs money on every run, which is why it pays to filter first. A filter node that drops duplicate variations or ones the model generated outside your industry cuts the bill without losing anything.

Classify by intent and dump it into a prioritized sheet

Once the data is back, the model comes in again, this time to classify. Each keyword is tagged according to what the searcher wants: to learn something, compare options, buy, or find a specific store or company. You also ask it to group into one topic the searches that a single page could answer, and to suggest the type of page that best fits each topic: a product page, a category or a blog post. Review that classification before trusting it, because a model can tag a search as commercial when in your industry it’s informational, and only someone who knows their customers will spot that.

The last node writes the result to another tab in Google Sheets, with one row per topic and these columns:

  • Main keyword
  • Secondary keywords
  • Total volume
  • Difficulty
  • Intent
  • Page type

If you sort that tab by volume and cross-check it with difficulty, you have a priority list you can move straight into your editorial calendar.

Schedule the run and get notified of new keywords

To make the workflow run on its own, the node that reads the sheet is started by a Schedule Trigger, the n8n node that launches the workflow at fixed intervals. In Make, the equivalent is scenario scheduling. For keyword research, once a month is usually enough, because search demand doesn’t change from one week to the next in most industries.

What’s really useful is hearing only about what’s new. n8n has a Remove Duplicates node with an option to remove items processed in previous runs, so that each month only keywords that have never come up before get through. If anything is left after that node, a Gmail or Slack node sends you the summary, and if there’s nothing, you don’t get a message.

For someone who has already used n8n or Make, setting all this up takes an afternoon or two. If you’re starting from scratch, allow a few days of trial and error, especially at the API step, which is where people get stuck most.

When automating keyword research isn’t worth it for an SME

We wouldn’t build this workflow for a neighborhood hair salon that lives off local customers. Its keyword list fits on a scrap of paper: “hair salon” plus the name of the neighborhood, a few variations with the type of cut, and not much else. Doing it once with ChatGPT and reviewing it once a year is all the keyword research it needs.

Nor is it worth it in local niches where the tools don’t record any volume. If you sell a highly specialized service in a specific area, the API will return zeros for almost everything, and you’ll be paying to confirm something you already knew. In those cases you get more out of conversations with customers, the questions that come in by phone or email, and Search Console reports, which show the real searches people already use to reach your website.

The third case is the most common. If your website publishes one article a month, or fewer, a workflow that suggests new topics every month will give you a list nobody is going to write. The bottleneck is content production, and automating the ideas part doesn’t move it.

Where we do recommend it is when the catalog is large or changes often, when you publish several times a week, or when you manage several websites at once. There, the time saved each month pays for the setup and the API bill.

Frequently asked questions about AI keyword research

Can it be done with just ChatGPT, without paying for any tool?

For generating ideas, yes, and for a small website it’s often enough. You ask it for the variations and the questions, sort them yourself using common sense, and write first about what your customers ask you. What you can’t do with ChatGPT alone is know which searches have demand, because the model doesn’t have that data.

How do I prioritize keywords without paying for a tool?

The most useful combination is ChatGPT or Claude for ideas, the Google Ads Keyword Planner to get a sense of the order of magnitude, and Google Search Console to see which searches already bring you visits. It’s slower and less accurate than a paid tool, but it’s usually enough to decide a blog’s next ten articles.

If you’d like us to build this workflow for your business, or first see whether it’s worth it for you, get in touch and we’ll look at it with you.

About the author

Francesc Sánchez — CEO and AI Consultant

I'm the founder and CEO of La Teva Web, latevaIA and Semseo Agency, and a consultant specializing in artificial intelligence, automation, GEO and vibe coding. With more than 24 years of experience in digital strategy and transformation, I help companies apply AI to optimize processes, improve their productivity and create new business opportunities. I'm also an expert advisor in digital transformation and AI accredited by ACCIÓ.

Articles by Francesc

Let's talk

Do you have a project in mind?

Tell us about it and let's grow your business together.

Analyze my sales opportunities