You find the questions your audience asks that your site does not answer. That is a content gap. Keyword tools find the ones with search volume attached. Language models find the ones phrased as questions, including the awkward ones nobody thought to type into a keyword tool. Used together they cover far more ground than either alone.
Most content calendars we see are built from whatever the team happened to think of that month. That works until you run out of obvious ideas, which usually happens around post fifteen.
Here is the workflow we would use to find the next fifty, and the parts of it that AI genuinely improves rather than just speeds up.
A content gap is a question your audience has that your site does not answer well, or at all. It comes in three shapes. Topics you have never covered, topics you covered badly, and topics you covered once and never updated. All three cost you traffic, and only the first one feels like a gap.
The second shape is the most common and the least noticed. A page that ranks at position 14 for a term you care about is not a missing page. It is a gap in depth.
The third shape is the sneakiest. Content that was correct two years ago and is now wrong reads fine to the person who wrote it and badly to everyone else.
Generate the questions that have no volume yet. Keyword tools report what people have already searched enough times to register. A language model can produce the long tail of ways a person might phrase a problem, including phrasings too new or too specific to appear in any database.
That matters more now than it used to. Answer engines expand a single query into many related ones behind the scenes, so the question you never targeted can still decide whether you get cited.
Models are also good at the reverse operation. Give one your existing article titles and ask what a reader would still be confused about after reading all of them. The answers tend to expose gaps in reasoning rather than gaps in keywords.
What they cannot do is tell you how many people care. That is still the keyword tool's job, and treating a model's output as demand data is the fastest way to write ten articles nobody wanted. Our post on topical authority covers why coverage and demand have to be balanced.
Start with the competitor comparison, because it is the most reliable input. Ahrefs' Content Gap tool exists to "find the keywords your competitors rank for, but you don't," and it lets you filter by keywords any competitor ranks for, keywords at least a set number rank for, or keywords all of them rank for.
That last filter is the useful one. A term every competitor ranks for and you do not is either a genuine gap or a deliberate choice, and either way you should know which.
Then export your own list of published titles and hand it to a model with a clear instruction. Ask it to group them into themes, name the themes that look thin, and list the questions a reader would still have. Give it your actual audience, because generic output comes from generic prompts.
Finally, cross reference. A gap that appears in both the competitor export and the model's question list is a strong candidate. One that appears only in the model output needs a demand check before it goes on the calendar.
It is the only source in this workflow that describes your actual site rather than the market. Search Console shows the queries you already appear for, including the ones you never targeted, and those near misses are the cheapest gaps to close.
Filter for queries where you have impressions but almost no clicks. Those are questions Google already associates with your site, on pages that are not answering them well enough to earn the click.
Then look at queries where a page ranks between roughly 8 and 20. Improving an existing page into that gap is usually faster than publishing a new one, because the page already has history.
Speed matters here. Ahrefs, in a study published in May 2025, found that only 1.74% of newly published pages reach Google's top 10 within a year. Editing something that already ranks avoids most of that wait, which we cover in our piece on how long SEO takes.
Ask the answer engines your own questions and read who they cite. Put your target question into ChatGPT, Perplexity, and Google AI Overviews, then note which sources each one pulls from. If the same three competitors appear every time and you never do, that is a citation gap.
Repeat it with variations. Answer engines fan a single question out into related ones, so asking the same thing five ways surfaces five different citation sets.
The shape of AI results also shifts fast, which is why this needs repeating rather than doing once. Semrush, analysing more than ten million keywords, recorded AI Overviews on 6.49% of queries in January 2025, 24.61% in July 2025, and 15.69% in November 2025.
The intent mix moved too. Semrush found that 91.3% of queries triggering an AI Overview in January 2025 were informational, and that by October the share had fallen to 57.1% as commercial and transactional queries grew.
It invents demand and it invents sources. A model asked for content gaps will happily produce a tidy list of plausible questions, some of which nobody has ever asked. It has no way to know the difference, because it is generating language, not measuring behaviour.
It will also attribute statistics to reports that do not exist if you let it. Any number that comes out of a content gap session has to be traced to a real source before it reaches a draft, which is the whole subject of our post on fact checking AI content.
The third failure is blandness. Models converge on the same obvious angles because those angles are what the training data contains. If your gap list looks like everyone else's, that is the model averaging rather than analysing.
Our practical guard is to treat model output as a list of hypotheses. Nothing goes on the calendar until a keyword tool, Search Console, or a real customer conversation supports it.
No, and Google has been explicit about it. Its guidance from February 2023 states that "Google's ranking systems aim to reward original, high-quality content that demonstrates qualities of what we call E-E-A-T: expertise, experience, authoritativeness, and trustworthiness." The focus is on quality rather than how content was produced.
Google draws the line at intent. Its post says that using automation to generate content "with the primary purpose of manipulating ranking in search results is a violation of our spam policies," and notes that SpamBrain continues regardless of how the spam was made.
It is equally clear the other way. Google states that "not all use of automation, including AI generation, is spam," and points to sports scores, weather forecasts, and transcripts as long standing examples of helpful automated content.
So using a model to find gaps is entirely safe. Using one to fill fifty of them overnight with unverified text is the thing Google is describing.
Score each one on three things: whether anyone is searching for it, whether you can say something true and specific about it, and whether it connects to work you actually sell. A gap that fails the second test is the one to drop, however good the search volume looks.
The second test is where most content strategies quietly fail. Writing about a topic you have no experience of produces a page that summarises other pages, and summaries are exactly what answer engines replace.
Prioritise depth over breadth when you are choosing. Ahrefs found that of the pages that did reach the top 10, 40.82% got there within a month, which suggests a strong page either lands quickly or does not land at all. Publishing more mediocre pages does not improve those odds.
We also weight gaps by how close they sit to an existing strong page. Filling a gap next to something already ranking builds a cluster, and clusters do more for authority than scattered posts do.
Run one competitor gap export, pull your Search Console queries for the last three months, and hand both to a model with your list of published titles. Ask it for the ten questions your site does not answer. Then verify each one against search volume before you write a word.
Expect a good share of that list to be useless. That is normal, and what survives the check is still more usable than a brainstorm usually produces.
Then fix an existing page before you write a new one. It is the fastest return available in this whole workflow and almost nobody does it first.
If you want help turning a gap list into a plan that fits what your business actually sells, we are happy to walk through it with you. You can find us at phoenix.studio, and we usually reply within a couple of days.
Tell us where you want to go. We'll tell you how we'd get you there.