Content Topic Research
SEO & AnalysisFreeNo account required
Crawls your website, catalogues what you have already published, maps the subject you operate in, and works out what is missing. Every topic it returns comes with a score you can inspect, an intent, a cluster, a recommendation to create, update, expand or merge, a cannibalisation check against your existing pages, internal links that are actually justified, and a research brief with an outline. Not a list of article ideas.
It reads your site before it suggests anything
The first thing it does is crawl your website and build an inventory: what every page is, what it is about, how deep it goes, what it links to. Nothing is recommended until that exists, because a topic suggestion made without knowing what you have already published is a guess with a confident tone.
Create is the last answer, not the first
Every topic is checked against every page you already have. A page whose subject is the topic gets an update. A page covering it as a section gets a spin-out with a link between the two. Two pages competing on it get merged. A new page is what is left when none of those apply, which on most sites is fewer topics than you would expect.
Clusters with an order, not tags
Opportunities are grouped into clusters, each with a pillar and its supporting pieces, and each cluster comes with the order to publish in and the reason for it. The pillar is the broadest topic rather than the highest-scoring one, because a cluster where six pages link up to a narrow case is a cluster built backwards.
Every number says where it came from
A topical gap is arithmetic over your own pages. Business relevance is a judgement. Competition is something read off live search results. Those are three different kinds of claim and the report labels each one, because a breakdown where all six numbers look alike is a breakdown implying six measurements.
Most candidates never reach you
Every scored topic passes a gate before it becomes a recommendation: is it relevant to what you sell, aimed at your readers, inside your subject, distinct from what you have already published, and worth the work. Topics that fail are discarded, counted by reason, and the count is shown. Forty topics worth writing beats four hundred that merely could be.
Confidence, separately from score
The score says how good an opportunity looks. Confidence says how much evidence sits behind that reading, and the two come apart constantly: a topic can score in the eighties on a partial crawl of a site whose business we could not establish. Each recommendation carries both, plus the observations it rests on and what limits them.
Research modes
Three research modes
The difference between them is what they are allowed to consult, and the tool is explicit about it in the interface as well as here.
| Mode | Live search | What it consults | What that means |
|---|---|---|---|
| AI Research | No | Your website, its existing content, the relationships between its topics, and AI analysis of all three. | Nothing in the output is a claim about what search engines show. The opportunity score is explicitly an AI-inferred assessment. |
| SERP Research | Yes | Everything above, plus two live search passes: what is being published on your subjects, and what people are asking about them. | Search-derived findings carry the queries that were run, the sources that were read, and the time of the search. |
| Deep Research | Yes | Everything above, plus a competitor-coverage pass and an entity pass, and a wider crawl. | The most complete picture, and the slowest. Competitor gaps only exist in this mode, because they need the competitor pass. |
How it works
What happens when you run it
Fifteen stages, and you watch each one finish. Nothing on the progress list ticks itself: every step appears when that work has actually completed, which on a large site takes a few minutes.
- 1
Your website is crawled
Up to forty pages, prioritising published content over utility pages and using your sitemap where you have one. robots.txt is honoured: disallowed paths are skipped and counted, because fetching dozens of pages nobody named is crawling by any reading of the standard.
- 2
Everything you have published is catalogued
Each page is classified by type, subject, search intent, business intent and depth, using its URL, its structured data and its own headings. A pricing page is not treated as an article. A blog index is not counted as coverage of a topic.
- 3
The subject you operate in is mapped
Not a map of your site: a map of the field, including branches you have never written about, because a map containing only what you already cover cannot show a gap. Then every branch is tested against your inventory. A branch counts as covered only when a substantial page treats it as its subject.
- 4
Candidates come from five places, four of them measurable
Phrases your articles keep using with no page of their own. Questions your headings ask without answering. Entities you name without explaining. Pages too thin to carry their own subject. Modifier expansion of what you sell. Then, and only then, the AI layer adds the adjacent subjects none of those imply.
- 5
Duplicates merge, but not the ones that only look alike
"Technical SEO Audit" and "Technical SEO Audit Guide" are one article under two titles and they merge. "How to Do a Technical SEO Audit" is a different piece that belongs underneath the first, so it stays separate with the relationship recorded. Collapsing all three loses a real opportunity; keeping all three creates a cannibalisation problem.
- 6
Then it is scored, gated, clustered and briefed
Eight factors, weighted with business relevance largest and topical gap second, and four more in the search modes. Then the gate: most candidates are discarded here, and the count is reported rather than hidden. What survives is clustered around the topical map, banded P1 to P4 with P1 held to one topic per cluster, and the strongest get a full research brief during the run.
Deliberate omissions
Six things this tool will not tell you
Every one of these appears in tools you have used, which is exactly why they are listed. If you go looking for one and cannot find it, it was left out on purpose.
Search volume
Volume comes from a keyword data provider. None is connected, so the report says “Data unavailable” rather than printing a number. A language model asked for a search volume will always produce one, because it has read thousands of articles quoting them, and that number would look exactly like a measured one to whoever plans a quarter around it.
Keyword difficulty, CPC, or a traffic estimate
Same provider, same absence, same refusal to estimate. Where a difficulty score would go, the report shows what it actually knows: how the field looked when it searched, and which sources it read to form that view.
Ranking positions
Even in the search modes. The search layer returns the sources it consulted, not an ordered result page, so there is no position in the data. Sorting those sources and calling the order a ranking would be inventing a metric out of a list.
A word-count target
"Aim for 2,000 words" is a guess wearing a number. Each brief describes the treatment the subject needs and why; how long that takes is the writer’s call, and a target the writer pads to reach makes the page worse.
Competitor topics presented as opportunities
A gap is only an opportunity if the topic suits your business. In the search modes the analysis is asked to say plainly when a competitor's topic suits their model and not yours, and those are listed as low-severity gaps rather than as things to write.
A long list
The quality gate discards most of what discovery finds, and the Research tab shows how many and on what grounds. A tool that hands you three hundred topics has moved the work of filtering onto you and called it thoroughness.
Who it's for
Built for people deciding what to publish
- Content leads deciding what a team writes next quarter, who need a reason for each choice they can put in front of a stakeholder
- SEOs inheriting a site and needing to know what it already covers before proposing anything
- Anyone whose content plan has produced two pages competing for the same query and wants to know where else that has happened
- Founders doing their own content who need to know whether to write something new or improve what is there
Scope and privacy
What it reads, and what it keeps
One crawl of your own site, at your explicit request.
What it reads
The HTML your server returns for up to forty pages of your site, plus robots.txt and your sitemap. It does not sign in, does not run your JavaScript, and does not submit forms. A page that renders its content in the browser is flagged in the report, because what we saw of it is what a crawler sees.
What it searches
In SERP Research and Deep Research, the subjects of your site are sent to Google Search through Gemini's search grounding, and the queries and sources are recorded in the report so you can see exactly what was searched. In AI Research no outbound search request is made at all.
What it keeps
Your research is stored in this browser, on this device, so you can come back to it. There is no account and no database: nothing about your site is kept on our servers after the run finishes. Clearing your browser's site data deletes the research, and up to six projects are kept before the oldest is dropped. Storage sits behind one interface in the code, so moving these projects onto a HastenOS account later will not change what the tool does.
Related
Once you know what to write
Content Topic Research decides what belongs on the site. When a page exists and needs to win a specific query, the optimizer reads that one page against that one query and says what to change.