Can ChatGPT Do Keyword Research? What It Gets Right and Where It Invents Things

ChatGPT is genuinely useful for keyword research and confidently wrong about it, often within the same reply. Ask it for topic ideas around your business and it will produce a solid, well-organised list in seconds. Ask it for search volumes for those keywords and it will produce an equally well-organised table of numbers it made up on the spot. The trick with ChatGPT keyword research is knowing which half of the output to trust, and this article draws that line clearly, with prompts you can copy and a free workflow for checking what it tells you.

What ChatGPT actually knows about keywords

ChatGPT has no connection to Google's keyword data. It has never seen a search volume figure for your niche, it does not know what people typed into Google last month, and it cannot tell you how competitive a term is. What it does have is an enormous amount of text about how people talk and ask questions, which turns out to be a large part of what keyword research is.

That distinction explains everything that follows. Language tasks: reliable. Data tasks: invented.

Where ChatGPT keyword research genuinely helps

Brainstorming topics from your business model

Describe your business in two sentences and ask for the questions your customers might search before buying. The output is usually broad, sensibly grouped and includes angles you would not have thought of alone. For a small business staring at a blank content plan, this is the fastest starting point available at any price.

Generating question and phrasing variants

People search the same problem in a dozen ways: "boiler making noise", "boiler banging sound", "why is my boiler loud at night". ChatGPT is excellent at producing these variants because predicting how people phrase things is precisely what it was trained to do. These longer, more specific phrases are usually the ones worth targeting first, for reasons covered in our piece on long-tail keywords.

Sorting keywords by intent

Paste in a rough list and ask it to split the terms into people researching, people comparing and people ready to buy. It does this well, and the grouping matters more than most owners realise: a blog post answers a research query, a service page answers a buying query, and mixing them up is one of the commonest reasons pages fail to rank.

Playing the sceptical customer

Ask it to role-play a customer who does not know your industry's jargon. An IT support firm says "endpoint protection"; the customer searches "stop staff clicking dodgy links". ChatGPT surfaces that gap quickly.

Where it invents things

Search volumes and difficulty scores

Ask ChatGPT for monthly search volumes and it will give you numbers. They look plausible, they come in tidy tables, and they are fabricated. It has no access to this data, so it generates figures that sound right for the shape of the question. The same applies to keyword difficulty scores, cost-per-click estimates and "trending" claims. It is not lying in any deliberate sense; it is completing a pattern, and the pattern includes numbers.

Local search terms

It knows far less about how people search in your specific town. Ask for keywords for a plumber in Chesterfield and you get generic plumbing terms with "Chesterfield" bolted on. It misses the neighbouring villages people actually search from, the local names for areas, and the terms that only make sense regionally. Google's own autocomplete, typed from your area, beats it comfortably here.

Freshness

Its knowledge has a cut-off date. New products, recent regulation changes and this year's terminology may simply be absent, and it will not always tell you so.

Prompts you can copy

These work in ChatGPT, Claude or any comparable assistant. Replace the bracketed parts.

• Topic seeds: "I run [business type] in [location] serving [customer type]. List 30 questions these customers might type into Google before they are ready to buy. Group them by the stage of the buying process."
• Variants: "Give me 15 different ways a non-expert might phrase this search: [keyword]. Include misspellings of technical terms and plain-English versions."
• Intent sorting: "Sort this keyword list into three groups: researching a problem, comparing options, ready to buy. Explain your reasoning for any borderline ones: [paste list]."
• Jargon translation: "Act as a customer with no industry knowledge who has this problem: [problem]. What would you actually type into Google?"
• Gap finding: "Here are the blog topics I have already covered: [list]. What obvious customer questions am I missing?"

Notice what is absent: no prompt asking for volumes, difficulty or "the best keywords to rank for". Those questions produce fiction.

A free ChatGPT keyword research workflow

Here is the full loop, using nothing paid.

1. Run the topic seeds prompt above and keep everything that sounds like a real customer question.
2. Type each promising phrase into Google and note what autocomplete suggests. Autocomplete is real search behaviour, free, and current.
3. Check the "People also ask" boxes on the results page for the same phrases. These are verified questions, not guesses.
4. Look at who currently ranks. If page one is national brands and Wikipedia, park the term. If it is small competitors and forum threads, it is winnable.
5. If you already have Search Console set up, check the queries you nearly rank for; those beat any brainstormed list because Google is telling you directly.
6. Feed the survivors back into ChatGPT with the intent-sorting prompt, then prioritise the buying-intent terms first.

To see this thinking applied end to end on a real business, we walked through a complete keyword research worked example rather than repeating it here.

When you still need a paid tool

Probably later than the tool vendors suggest. A small business publishing a few posts a month can run entirely on the free workflow above. Paid tools earn their fee when you need to compare hundreds of keywords at once, track rankings across many pages, or research a market you know nothing about. If you are only choosing your next ten article topics, verified phrasing plus a look at who currently ranks answers the question. When you do have a long list and need to order it, prioritising is its own skill, and it matters more than the tool that produced the list.

Where Helio SEO fits in

Helio SEO does this research automatically: it scans your site, works out who your buyers are, builds a keyword plan and shows its reasoning in the open, then writes and publishes an article a day against it. You can read how the process works if that appeals. If you enjoy the research and have the hours, the workflow above needs nothing more than ChatGPT and a browser.