AI pricing keyword trends

The keyword cloud

Topics by category

The 45 topics split into 6 categories. Model pricing dominates, reflecting how much of current AI-related search is people trying to work out what a model actually costs to use, whether by API or subscription. For the actual verified prices behind these topics, see the model release tracker and subscription-vs-API calculator.

How to read this, and what it is not

This is a frequency signal, not a volume figure. Search autocomplete returns the suggestions a search engine considers most likely to complete what you have started typing, which correlates with how often people search a phrase but is not a measurement of it. A topic that rolled up many suggestion variants is one the engine expects a lot of different people to be reaching for. It is not a claim that the topic gets a particular number of searches per month, and it should never be quoted as one. An advertising platform’s keyword planner is the tool for that question, and it will disagree with this page.

Three biases are worth holding in mind. Autocomplete is personalised and localised, so a collection taken from one place at one time is a slice rather than a census. It skews toward phrases people type rather than speak, and toward the beginnings of questions, so long or unusually specific queries are systematically under-represented. And it reflects what people are asking now, which means a term spikes when a model launches and decays afterwards, whether or not the underlying interest is durable.

What it is genuinely good for is shape rather than magnitude: seeing which pricing questions cluster together, noticing that a comparison people ask about constantly has no good answer written anywhere, and catching a term entering the vocabulary before any keyword tool has enough history to report it. Read it as a map of what the market is confused about, then check any specific term against a volume source before acting on it.

The collection is rebuilt rather than accumulated, so each refresh replaces the previous reading instead of averaging into it. The as-of date at the top of the page is the date the suggestions were pulled, and comparisons between two refreshes are comparisons between two snapshots, not a trend line.

For the coding-agent and hardware topics mapped here, see the Claude Skills and agent directory, cost-per-task ranking, inference providers directory, and DDR5 price tracker.