Sometimes the topic hunt hands you a topic that is obviously wrong. Not subtly off, not a coin-flip call, but visibly bad. A B2B SaaS blog gets pitched a listicle for consumers. A niche developer tool gets a topic that would suit a shampoo brand. A serious explainer comes back pitched at someone who would never buy anything.
The temptation is to decline the topic, run the hunt again, and move on. Sometimes that works. Often the same shape of bad idea comes back on the next run, because the actual problem is not the topic. It is upstream, in the website profile.
What the topic finder actually reads
This is worth being precise about, because it decides where the bug can be.
A topic hunt hands the agent your website address, the project's writing rules, the website profile, the titles of the posts the project already has (so it does not suggest the same thing twice), the topics you declined with the reasons you gave, and, when you started the hunt yourself, the focus you typed. That is the whole input. No search volume, difficulty score, ranking data or results page reaches the topic finder, so no suggestion can be standing on one. The agent may look at your site to see what is already covered, but the document it is told to treat as established fact about the business is the profile.
It gives back five topics, each with a title and an outline of two or three sentences, and each optionally carrying up to four questions: the facts it could not find out on its own.
So when a suggestion is wrong about your business, it is almost always because the profile is wrong about your business, or vague enough to be read the wrong way.
Start from the topic and read backward
Do not start at the top of the profile. Start at the topic the agent proposed, then follow the trail up.
The suggestion does not come with a chain of reasoning. RankMill keeps no record of the steps an agent took, and the Activity tab shows a run's status and timing, not its thinking. So the outline plus the profile is the whole debugging surface. Read the outline first: who is it written for, what shape of post is it, what does it assume the business sells? Then open the profile and find the sentence the topic is standing on.
Research writes the profile in eight sections, in the same order: Business, Products & services, Audience, Voice & tone, Positioning vs competitors, Terminology & things to avoid, Competitors analysed, and Topic opportunities. Most bad topics break in one of three places. Once you know the shape of each break, you can spot it in ten seconds.
Failure mode 1: the business is described too loosely
This is the most common. The Business and Products & services sections are true but general, and a general description has a much wider set of plausible topics hanging off it than a specific one.
Signals in the topic and the profile:
- The outline is about your category rather than your product
- The topic would suit any of five companies in your space equally well
- The profile describes what the product is rather than what it is for
- The profile never says who the product is not for
A loose profile usually shows up as a topic that feels adjacent to your niche but not inside it. If your product is a Postgres monitoring tool and the profile calls it a database tool, a piece on choosing a database GUI is a reasonable read of that sentence. The agent is not wandering. It is following a sentence you can go and read.
The fix is to make the profile specific. Edit Business and Products & services to say what the product does, who it is for, and in a line, who it is not for, in words that rule out the adjacent niche. The Profile tab is the editor, so you can change the text in place and it saves itself, or ask the AI for the change and read the profile as it would read before you accept it. Every later hunt reads the corrected version, because each run reads the profile live.
Failure mode 2: the Audience section names the wrong reader
The business is described correctly and the topic is a real topic. It is simply written for someone who is not going to buy from you.
Signals in the topic:
- The outline is pitched at a beginner when you sell to specialists, or the reverse
- The post would be read by the person who uses the product rather than the person who chooses it, when those are different people
- The outline assumes a budget, a team size or a level of authority your readers do not have
- The topic is right for the audience you had two years ago
This one is sneaky, because nothing in the profile looks false. The Audience section says something like "developers and technical teams", which is true of almost everyone who has ever visited your site, and is therefore no constraint at all.
The fix is to write the Audience section as a person with a job, a problem and a decision to make, not as a demographic. Name the role that signs off. Name the moment they come looking. If two audiences matter, say which one the blog is for and what the other one gets instead. A sharper Audience section changes the angle of every suggestion on the next run, not just the one that annoyed you.
Failure mode 3: the wrong peers under Positioning vs competitors
The Positioning vs competitors and Competitors analysed sections are where the agent learns what it is differentiating from. Get the peers wrong and every suggestion inherits the mistake, because a good suggestion is defined partly as one your competitors do not already cover.
Signals in the profile:
- The competitors named are companies that share your vocabulary but sell to a different buyer
- They include publishers and directories, which cover your category without competing for your customer
- The peer you would name first in a sales call is missing
- The positioning says what you have in common with them rather than what separates you from them
Wrong peers produce topics that are defensible in isolation and wrong for you. If the peers are mostly media sites writing broad primers, the differentiation the agent reaches for is a broader primer of its own, which is the exact wrong direction for a product blog that usually wins by going narrower than the coverage around it.
The fix is to constrain the competitors to true peers and to say, in a sentence under Positioning vs competitors, what you do that they do not. If a name under Competitors analysed is a publisher rather than a competitor, either take it off or say what it is, so the agent stops treating its coverage as the bar to clear.
When to fix the profile and when to skip the topic
One-off odd topics happen. Five suggestions come back at once and one of them being weak is not a bug worth chasing. Decline it, with a reason if you have one, and move on: every later hunt is told the declined title and the reason, so that topic should not come back.
If the same shape of bad topic appears across runs, it is standing on the same profile sentence each time. That is the signal to edit. A correction you make once removes a whole category of future noise, which is worth more than skipping one topic at a time.
And the profile is worth rereading on good weeks too, not only during a fire drill. Every hunt reads it, and knowing what it says when the topics are fine makes the sentence behind a bad one jump out. A useful habit for a content operator: glance at the latest batch, pick the topic that surprised you, and find the profile sentence it came from. If nothing surprised you, do the same for the one you liked least. That is usually where the interesting failure lives.
The short version
- The topic finder reads your site address, the writing rules, the profile, your existing titles, your declined topics and any focus you gave. No search data
- So read backward from the bad topic to the profile, not forward from a query list
- Loose business description: say what the product is for, and who it is not for
- Wrong reader: write the Audience section as a person with a decision to make
- Wrong peers: cut the publishers and say what separates you from the real ones
- Decline one-off failures, edit the profile on repeat failures
- Reread the profile on good weeks too, so the bad sentence is easier to spot
The topic finder is not guessing in a black box. Every proposal it makes is built on a document you can read, edit and correct, and every correction is a chance to make the next hunt better.