GEO Topic Selection: How to Find the Content Gaps AI Still Answers Poorly
The rule for picking a GEO topic is to find questions your customers ask often, that AI still answers poorly, and that you know best. Count how many different websites AI cites to judge scarcity, ask AI the question once to find what it misses, blurs or gets wrong, then add the chosen question to your measurement list.
Also available in：繁體中文
The rule for choosing a GEO topic is to find questions your customers ask often, that AI still answers poorly, and that you happen to know best. GEO topic selection takes two checks: ask a few AI tools with web search about each candidate topic and count how many different websites they cite, which tells you how scarce sources are; then read the current AI answer and list what it misses, blurs or gets wrong. Each of those is a content gap, and the content gap is what your article must fill.
Topic selection is Step 2 of the five GEO steps, after measuring and before writing. This article covers where candidate topics come from, how to judge scarcity and content gaps, how to narrow a topic that is too broad, and what to do once you have chosen.
Why does topic choice matter more than writing technique?
A well-written article on a topic that large sites already dominate may never get cited. The same article on a topic almost nobody covers can become the main source AI relies on.
Before answering, AI searches the web and picks passages from the pages it reads (Google Search Central, Perplexity docs). The fewer pages there are to choose from, the better the chance yours is picked.
Small sites have a real chance. In the GEO paper's experiments, adding cited sources raised the share of the answer for a site ranked 5th in search results by 115.1%, while the site ranked 1st dropped by 30.3% (GEO paper, Table 2). AI cares whether content is easy to cite, not only how big the site is.
The three conditions of a good topic
A topic worth writing first meets all three conditions:
| Condition | Question to ask yourself | How to check |
|---|---|---|
| Customers ask it | Would customers really ask AI this question? | Questions that keep coming up in support, sales or comments |
| AI answers it poorly | What is the current AI answer missing or getting wrong? | Count cited websites and ask AI the question once |
| You know it best | Do you have numbers, steps or experience others lack? | You can write concrete prices, steps, cases and caveats |
Meet only the first two and you cannot write something better than what exists. Meet only the last two and nobody is asking. A topic that meets all three is the first choice for GEO topic selection.
Where do candidate topics come from?
You do not have to invent candidate topics. These three places are the fastest:
- Measurement results: questions where neither you nor any competitor was mentioned, and questions where AI mentioned you but cited someone else's page. See how to measure AI visibility.
- Questions customers actually asked: support logs, questions sales hears often, on-site search, social comments. Write them down in the customer's own words.
- Things you explain to customers again and again: processes, prices and rules you have to repeat every time are usually what nobody online has explained clearly.
Collect 10 to 20 candidate topics first, then filter them with the two methods below.
Method 1: Judge scarcity by counting the websites AI cites
Scarcity is how many existing sources a topic has. Turn each candidate topic into a question, ask a few AI tools that search the web, such as ChatGPT search, Perplexity and Google AI Overviews, and count how many different websites appear in their cited sources. A website cited several times counts once.
| Websites cited | What it means | What to do |
|---|---|---|
| Almost none | Almost nobody writes about it, so AI has to guess | Write it first; the opportunity is largest |
| A few | Some coverage, but few choices | A more complete, better-sourced article can rise to the top |
| Many | Heavy competition; large sites may already own it | Take a narrower angle, or come back later |
Take a bookkeeping firm as an example (illustrative numbers): for "what is business bookkeeping", AI cited 9 websites, mostly large accounting platforms; for "what should a new Taipei startup watch for in its first year of bookkeeping", it cited only 2. The second is far scarcer and a better GEO topic.
Make sure web search is turned on. An answer without web search relies only on training data, lists no sources, and cannot show you scarcity.
Method 2: Read how AI answers now to find content gaps
Write the topic as a complete question, ask AI once, and sort the answer into two groups:
- Points it already gets right: your article should cover them too, but they are not the focus.
- What it misses, blurs or gets wrong: these are the content gaps, and they are what your article must fill.
There are four common kinds of content gap, and usually only people who actually do the work can fill them:
| Content gap | What AI often says | What you can add |
|---|---|---|
| Concrete numbers and prices | "Costs vary by situation" | Real price ranges, how pricing works, and units |
| The actual order of steps | A few terms with no sequence | What comes first, which documents to prepare, how long each takes |
| Regional or regulatory differences | General rules only, or another country's rules | Your local rules, with a link to the official source |
| Practical caveats | Nothing, or a generic reminder | The mistakes customers make most, real cases you handled |
The more specific content gaps you find, the better the article's chance of being cited. Places where AI gets the answer wrong are especially worth writing about, because your article can become the source that corrects it.
How do you narrow a topic that is too broad?
When scarcity shows "many", you do not have to give up. Narrow the topic by adding a region, an audience, a situation or a time frame:
| Approach | Topic too broad | Narrowed topic |
|---|---|---|
| Add a region | Bookkeeping fees | How much bookkeeping costs per month for a small company in Taipei |
| Add an audience | Tax filing tips | What a one-person company should watch for in its first business tax filing |
| Add a situation | Best bookkeeping software | Do you still need bookkeeping software after outsourcing your books? |
| Add a time frame | How to register a company | Which filings are due within 30 days of registering a company |
Fewer people search for a narrowed topic, but those who ask are usually closer to a decision, and large sites rarely go into that much detail.
How do you set priorities?
After filtering, decide which article to write first in this order:
- Topics with high scarcity, many content gaps and strong expertise on your side come first.
- Questions where measurement showed AI mentions you but does not cite you come second; AI already knows you and only lacks a citable page.
- Topics with medium scarcity where you have exclusive numbers or cases come third.
- Broad, competitive topics get narrowed first; those you cannot narrow wait until your site carries more weight.
Choose only 1 to 3 topics at a time. Choose too many and none of them gets written in depth.
What do you do after choosing a topic?
Once you choose a topic, add its question to your Step 1 measurement list. Only then can you see whether AI's answer to that question changes after you publish.
Next, turn the content gaps you found into a list and use it as the skeleton of your outline; for how to write it, see 5 tips for AI citation. Research shows that adding verifiable sources, concrete statistics and quotations raises a page's visibility in AI answers by about 30–40% relative to the original (GEO paper), so every number that fills a gap should carry its source.
For all five steps, see the GEO overview; for how AI picks the pages it cites, see how AI chooses its sources.
Frequently Asked Questions
Q: What is GEO topic selection?
GEO topic selection is finding the topics most worth writing and most likely to be cited by AI.
A good topic meets three conditions: customers ask it often, AI still answers it poorly, and you know it best. You judge it by scarcity and content gaps.
Q: What is a content gap?
A content gap is the part of AI's current answer that is missing, vague or wrong.
Common content gaps are concrete numbers and prices, the actual order of steps, regional or regulatory differences, and practical caveats. Those are what your article should fill.
Q: How do I judge a topic's scarcity?
Turn the topic into a question, ask a few AI tools with web search, and count how many different websites they cite.
A website cited several times counts once. Almost no citations means write it first; many citations means heavy competition, so narrow it down.
Q: Why can small sites get cited by AI?
Because AI cares whether content is easy to cite, not only how big the site is.
In the GEO paper's experiments, adding cited sources raised the answer share of a site ranked 5th in search by 115.1%, while the 1st-ranked site dropped by 30.3%.
Q: Where do I find candidate topics?
From measurement results, questions customers actually asked, and things you explain to customers most often.
Measurement questions where nobody was mentioned, or where you were mentioned but not cited, are the most direct. Support logs and sales questions are closest to real demand.
Q: What if a topic is too popular and competitive?
Narrow it by adding a region, an audience, a situation or a time frame.
For example, turn "bookkeeping fees" into "how much bookkeeping costs per month for a small company in Taipei". Large sites rarely cover that, and the people asking are closer to a decision.
Q: How many topics should I choose at once?
One to three is enough.
Choose too many and none gets written in depth, so the content gaps stay unfilled. Write, publish and measure first, then pick the next batch.
Q: What is the first thing to do after choosing a topic?
Add the topic's question to your measurement list.
That lets you compare AI's answer to that question before and after publishing. Then list the content gaps you found and use them as the skeleton of your outline.
Start monitoring your brand's AI visibility
Free sign-up, 5-minute setup, an automatic BVS report every week
Start free