Using AI for Content Ideas Without Sounding Like Everyone Else
Ask any assistant for blog ideas for your industry and you get the same list your competitors got. That is not a flaw in the tool. It is a consequence of what you gave it.
Why the first ideas are always generic
These models produce what is most probable given the prompt. A generic prompt describes thousands of businesses, so the output describes thousands of businesses. Blog ideas for an HVAC company returns something close to the average of every heating and cooling blog ever written, because that is literally what it is.
The specificity has to come from you. That is not a workaround, it is the whole technique.
Give it something only you have
Four sources, all free, all already in your business:
Your reviews. Complaints and praise, in your customers’ own words.
Your sales and service calls. The questions people ask before they commit.
Your support email. The confusions that repeat until you stop noticing them.
Your Search Console queries. What people actually typed before they arrived.
Paste a batch in and ask for recurring themes ranked by frequency. The output is grounded in your business and nobody else’s, and there is nothing for the model to invent because you supplied every fact.
Three prompt patterns that produce specific ideas
The objection pattern
Here are forty real objections from sales calls. Group them, and for each group propose one article that would answer it before the call happens.
The sceptic pattern
Write the three questions a sceptical buyer would ask about this claim. The questions are almost always better articles than the claim was.
The constraint pattern
These are the six municipalities we serve and the three services we offer. Which combinations have a genuine difference worth writing about, and which would produce near-identical pages? The second half of that question is the valuable half.
A Michigan example
As an illustration, take a landscaping company near Grand Rapids. The generic prompt produces Five Lawn Care Tips for Spring, which already exists several thousand times.
Suppose that, fed its own review text, the recurring complaint turns out not to be lawn quality at all but communication about rescheduling after rain, which in a Michigan spring happens constantly. The article writes itself: how weather rescheduling works and what customers can expect. That piece also happens to answer the most common support call of the season.
One of those is filler. The other reduces inbound calls and converts, and no competitor will publish it because no competitor has that review data.
The editing pass that matters
Three questions against any idea before it reaches a calendar:
Could a competitor publish this unchanged? If yes, it is not yours.
Does it contain anything only we know? If no, it will not earn attention.
Would the person who asked us this actually read it? If no, it was written for a search engine.
Discarding most of a list is the normal outcome and a sign the filter is working.
What to never let it invent
Prices, specifications, timelines or guarantees.
Local facts: neighbourhoods, landmarks, regulations, service areas.
Claims about results, yours or anyone else’s.
The test is simple: anything a reader could act on and be wrong about is yours to verify, not the tool’s to supply.
The full guide
Content ideation is one technique among several. The complete guide covers research, paid variation, email, reporting and the B2B reality most AI advice ignores.