Better prompts won’t fix an AI that’s missing the research
Your AI assistant writes a competitor report. It sounds fine. It could have been written six months ago. It lists broad trends, names no specific posts, and doesn’t tell you what actually changed this week.
Adology’s team argues the problem isn’t your prompt. It’s what the prompt doesn’t include.
What a Brief Leaves Out
A brief tells the AI what the job is. It rarely includes the facts behind the job. For a competitor report, those missing facts often include the specific brands to compare, the accounts and channels to check, the date range, the baseline from prior weeks, legal-approved brand claims, and the reasons the team rejected a direction last time.
An experienced strategist knows all of this without writing it down. The AI doesn’t gain it from a well-crafted prompt. The team has to supply it directly.
Specific Requests Beat Open-Ended Ones
The Adology post contrasts two requests. The first: “Summarize what furniture brands are doing on social media.” The second: “Review these six furniture brands from the past seven days. Compare their paid ads and regular posts with the previous four weeks. Show new offers, repeated formats, and changes in creator partners. Link to the posts behind each finding. Flag any finding based on missing data.”
The second request is useful because it names the brands, dates, content types, and the proof required. It gives the AI less room to return a broad answer that doesn’t help with Monday’s report.
Keep the Standing Rules Somewhere Permanent
Report dates change. Most of the other rules don’t. The competitor list, the method for separating paid from organic posts, legal limits, and brand fit criteria are the same week to week. If a strategist has to re-paste all of this into a new chat each Monday, details drift. One person uses five competitors, another uses eight. One treats a boosted post as organic.
The recommendation is to maintain a shared document with the approved competitor list, the accounts to include, brand voice and legal rules, how the team interprets engagement metrics, and past decisions with their reasoning.
Ask for Output a Human Can Actually Check
“Find insights” gives a reviewer nothing to judge. A more useful request: five specific changes, a link to the source behind each one, and one suggested next step. Now the reviewer can confirm whether each change is genuinely new, open the source, and decide whether the next step fits the brand.
Corrections should feed back into the setup. If the reviewer removes the wrong competitor or reclassifies a paid post, that choice should be part of next week’s rules, not buried in a one-off chat.
The Operator Test
Four questions that tell you whether an AI workflow is actually working:
- Did it find changes the reviewer agrees are new?
- Can the reviewer check the source behind each finding?
- Can the team correct the report without starting over?
- Will those corrections carry into the next run?
If all four answers are yes, the workflow is accountable. If not, a longer prompt won’t fix it.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
