Topic-first SEO means you start with the subjects your customers are deciding on, and only then look at keywords, competitors and pages. Keyword-first SEO starts with a list of search terms sorted by volume and works backwards from there. The order looks like a detail. It decides what you publish, how many pages you need, and whether Google and the AI answers have a reason to send you visitors.
I run my own websites this way, and I built SitePerfector around it, so I have some opinions on this.
What changed in search
Google groups searches by meaning. Search "best espresso machine", "best home espresso machine" and "best espresso machine for home". You get almost the same ten results. To Google those are one question, asked three ways. The old niche-site playbook assumed the opposite: one page per exact search term, the wording matched to the letter, dozens of near-identical pages per site (good times, while they lasted). It stopped working, and you can watch why in SitePerfector's clustering step: it reads the live results for every keyword in a market, and keywords that mean the same thing come back with almost the same ten results, over and over. Google's own guidance says to write for people, not for keyword variations (Google, creating helpful content (opens in new tab)).
AI answers cite their sources. ChatGPT, Gemini and Google's AI Overviews now answer first and link to where the answer came from. They do not cite twelve pages that each cover a slice. They cite the page that answers the question properly.
So a site built that way now has five pages competing for one result, and none of the five is complete enough for an AI answer to cite.
Keyword-first SEO, and why it stopped working
You know the routine. Export a keyword list. Sort by volume. Filter by difficulty. Pick the winners. Write a page for each. Repeat next month.
Three things go wrong.
Your pages fight each other. Five variations of one question become five thin pages. Google picks one, usually not the one you wanted, and the other four split the signal.
You get data, not a plan. A list of 800 keywords tells you what people type. It does not tell you what to build, in which order, or when you are done. So you spend the hours sorting, and the sorting still does not produce a decision.
The numbers are thinner than they look. The difficulty score in every keyword tool is bought from a data vendor, and the vendor mostly has no data for the long tail. I tested it on 400 long-tail keywords from one niche. A score came back for 8 of them. Two percent. The other 392 were blank, and most tools fill a blank with a default number and show it to you as if it were measured.
Example. "harbor freight air compressor" gets just under 50,000 searches a month. The difficulty score for it comes back as zero. That does not mean easy. It means the vendor has no data for it. Type it into your keyword tool and look at the difficulty column. Then ask how many of your smaller keywords are showing you a real number.
Difficulty coverage: 400 long-tail keywords, one niche
Topic-first SEO, defined
I like to think of a topic as 'one thing your customer is deciding': which espresso machine to buy, whether to repair or replace a compressor, how to price a coaching package. A topic is bigger than a keyword and smaller than your business.
In SEO land the structure has four levels. Tools and guides use different names for them (pillars, silos, hubs), but it is the same four levels underneath. These are the names I use, and the ones you see in SitePerfector:
| Level | What it is | Espresso example |
|---|---|---|
| Theme | An area of your business | Home espresso |
| Topic | One decision your customer makes | Choosing a home espresso machine |
| Cluster | Keywords whose search results overlap. They belong on one page | best espresso machine, best home espresso machine, espresso machine reviews |
| Page | One question of the topic, answered completely | Manual, semi or fully automatic: which one fits you? |
The rule: keywords with the same results share a page. Questions with different results get their own page. That one rule replaces most of the sorting. It is also the step SitePerfector does for you: it reads the live top results for every keyword in your market, roughly 40,000 results per map, and groups keywords by how much their results overlap. Out comes a short list of pages, around a hundred, each with the keywords it covers underneath. That is a lot easier to work with than a list of 1,500 keywords, because every line is something you could build.
The two methods side by side:
| Keyword-first | Topic-first | |
|---|---|---|
| You start with | A keyword list sorted by volume | One decision your customer makes |
| You publish | One page per keyword variation | One page per real question |
| Your pages fight each other | Yes, by design | No, by design |
| You know when you're done | Never | When the topic's questions are answered |
| What Google sees | Five versions of one answer | A site that knows the subject |
| What AI answers can cite | Nothing deep enough | The page that answers the question |
Why topic-first SEO works
I see four reasons.
One clear signal per page. Google judges pages one at a time and can rank a single passage on its own. A page that answers one question is a clean unit. A 4,000-word guide where your answer sits in section seven is a weak one.
Coverage reads as expertise. A site with a page on every question inside a topic, linked to each other, looks like it knows the subject. Twelve overlapping pages look like a content farm. Google's starter guide tells you to group related content together (SEO starter guide (opens in new tab)), and the difference shows in what it sends you.
AI answers need a quotable unit. An AI answer takes a passage that answers the question in plain words, from a page whose heading matches the question. That is exactly what a topic-first page is. There is no separate trick for AI. The same page serves both.
Less work, and it ends. Keyword-first never finishes. There is always another variation. A topic has a finite set of questions: four to eight pages, then you move to the next topic. Fewer pages, and better ones.
How to do it, step by step
There are three steps, and each one links to the Academy page with the detail.
Step 1: Find your topics
- Write down the decisions your customers make before they buy from you, and the questions they ask before, during and after. Ten to twenty. That list is worth more than any export.
- Check demand with a keyword tool, but use it as a yes or no: do people search for this, and roughly how many. Do not sort by it. Keyword tools have very little data on long-tail and bottom-of-funnel searches, which are the ones that bring buyers, so a blank or a tiny number there usually means the tool has no data. It does not mean nobody searches for it. Chasing the highest-volume keywords is a habit to let go of.
- Keep the topics close to what you sell. A topic can have great numbers and bring you nobody who buys.
Detail: How to find topics that bring the right people.
Step 2: Build your topic map
- For each topic, list the questions. Search each one and look at the top ten.
- If two questions return mostly the same results, they are one page. If the results differ, they are two pages.
- Look at who ranks. Sites your size or smaller: you can compete now. Only giants: build up to it, or leave it. Write that call next to each topic.
- Order the topics: closest to money first, easiest to win second.
A small site beats bigger ones on specific questions. On one niche map I measured, a site with roughly 8,600 monthly organic visits out-ranks domains with a median of about 15,800 visits, and beats sites above 400,000 visits on some long-tail questions. The big sites win the short, broad searches like "espresso machine", because those come down to overall authority. On a specific question, the page that answers it best wins, and a small site can write that page. So when you look at who ranks for a topic, ask whether they are bigger than you on this question. Their total traffic tells you little.
Worked example: one niche map, roughly 40,000 ranked positions compared
Detail: Your topic map.
Step 3: Write pages that get quoted
- One question per page. The heading is the question, in the words people use.
- Answer in the first two sentences. Then explain.
- Add something only you know: a number from your own work, what goes wrong in practice, where you disagree with the usual advice. Without it the page is copy number 1,001 of the same answer.
- Keep it to the question. 600 to 1,200 words is usually enough. Link to the other pages of the topic.
Detail: Writing for citation and Writing with AI without sounding like AI.
What to do this week
- Pick one topic: one decision your customers make. Not your biggest one. The one you know best.
- List its questions. Six to ten.
- Search each one. Merge the ones with the same results. You should be left with four to eight pages.
- Write the one page you can add the most to. Answer first, one question, something only you know.
- Publish it, link it from the most related page you already have, and start the next one.
That is topic-first SEO, by hand. What matters is the order: topic first, keywords second, pages last.
Where SitePerfector fits
I built SitePerfector to do step 1 (find your topics) and step 2 (build your topic map) with real data instead of guesswork. It vets up to 1,500 keyword ideas, reads roughly 40,000 live search results, maps every competitor by topic, and gives each topic a call: good to start, or wait. Then it carries the plan into step 3 (write pages that get quoted) with a blueprint per page. If you would rather do it by hand, the Academy shows every step, free.