TL;DR: Start with one instruction on one account: require every AI answer to show its role, its sources, any conflicts, a confidence score, how long the answer stays valid, and the decision it hands back to you. Structure turns a fluent answer into one you can check, though no prompt can make the AI accountable for the result.
What is structured prompting?
Structured prompting means asking an AI for a defined shape of answer every time, so the parts you need to judge are always in the same place, and the aim is an answer a person can verify.
Anthropic starts from the same idea of defined success. Its prompt engineering overview lists, under “Before prompt engineering,” what it assumes a user already has: “A clear definition of the success criteria for your use case,” “Some ways to empirically test against those criteria,” and “A first draft prompt you want to improve.” The prompt comes third in that list, after the measure and a way to test against it.
Basil C. Puglisi, MPA, built his approach on the same order of operations. His HAIA-RECCLIN methodology starts with RECCLIN Reasoning, a ten-field output format that makes an AI “show its work, cite its sources, score its own confidence, flag its own conflicts, and hand the final choice back to a person,” so the prompt becomes a standing requirement that every answer has to meet. The page also separates the method from the usual meaning of the term: “A better prompt improves what one AI returns to you. RECCLIN changes who is accountable for what happens next.”
What should every answer show?
Every answer should show enough for a person to decide whether to trust it, and the RECCLIN Reasoning format sets that floor at ten fields. In the page’s words, “Ask any AI to answer in the RECCLIN format and every response comes back carrying these ten fields”:
- Role and Task, which reveal how the AI read the request before a misunderstanding spreads.
- Output, the answer itself.
- Sources, “Citations you can check.”
- Conflicts, where “Finding none is a claim that has to be stated, not a blank to skip.”
- Confidence, “A score from 0 to 100 with the reasoning behind the number.”
- Expiry, because time-sensitive information treated as permanent is a risk.
- Fact, Tactic, KPI, the evidence, the action it points to, and the measure that would show it worked.
- Recommendation and Decision, the AI’s advice kept separate from the evidence, and the choice handed back to you as options.
The fields cost time on purpose. In the format’s own words, “The format trains the person before it standardizes the output,” and “A reader who has learned to check a Sources field and question a Confidence score has stopped accepting AI output on tone alone.” The page puts the cost at fifteen to thirty minutes of review per session in the first month, dropping to five or ten minutes once reading the fields becomes routine.
The fields also matter more as model reasoning becomes harder to see. The page notes that “The fields never showed the reasoning.” What they do is require the model to declare what it used and what it disagreed with, along with its confidence, and the page argues that as provider reasoning traces close or turn into summaries, that declaration becomes the main account a model still gives an ordinary operator.
How do you set it up on one account?
Setting it up costs nothing, because, as the HAIA-RECCLIN page puts it, “RECCLIN Reasoning runs on one AI platform at any tier, including free accounts, and the loading prompt is published in the paper.” The full RECCLIN paper on reasoning and dispatch carries the prompt, and Basil’s AI personalization prompts page carries copy-ready custom instructions for each major platform.
Load the instruction into the platform’s custom instructions or memory so it applies to every session without anyone having to remember to paste it. Then test it. The personalization page suggests a single test question, “What are the top 3 risks in [your industry] for Q1 2026?”, followed by a short checklist: the AI should declare its role, provide sources, note conflicts or state “None identified,” include a confidence level, and end with a decision point that asks for your approval. For a current test, swap in the coming quarter. Run the test after any platform update, since a changed model can drop fields without warning.
The role list behind the format appears in Appendix A, “HAIA-RECCLIN Framework Definition,” of Basil’s book Governing AI: When Capability Exceeds Control, which defines the HAIA-RECCLIN framework and its seven roles: Researcher, Editor, Coder, Calculator, Liaison, Ideator, and Navigator.
What can a prompt not do?
A prompt cannot make an AI obey it. Basil’s account of why you cannot program or prompt governance into AI describes a June 2026 session in which the instructions to pause for authorization and to retrieve sources were present, and “completion pressure displaced both.” The conclusion is direct: “a prompt is a request, not a control.”
That limit shapes how to use structure. When a model follows the format, declaring its sources and handing back the decision, any point where it substituted its own judgment is easier to spot in the fields, though nothing in the prompt forces the model to follow the format. The decision itself still belongs to a person, and Puglisi Consulting’s AI Literacy and Workforce Review asks whether staff “direct, challenge, verify, and own the work they do with AI,” which is the skill structured prompting supports but cannot replace.
Questions Readers Ask
Is structured prompting the same as prompt engineering?
The two overlap, and the target differs. Prompt engineering usually aims at a better answer from one model, while structured prompting in the RECCLIN sense aims at an answer a person can evaluate, with sources to check, conflicts stated, confidence scored, and the decision handed back. The HAIA-RECCLIN page frames the difference as a change in “who is accountable for what happens next.”
Does it take several AI platforms to start?
No, because RECCLIN Reasoning runs on one platform at any tier, including free accounts. Several platforms help later, when comparing answers reveals disagreements a single model cannot detect, and the personalization guide’s quick setup checklist calls for three platforms for that multi-AI setup. One account with the instruction loaded is enough to build the habit of reading the fields.
Will the ten fields slow down the work?
Yes, at first. By the HAIA-RECCLIN page’s estimate, the first month adds fifteen to thirty minutes of review per session, and that falls to five or ten minutes once reading the fields becomes routine. The slowdown is the point, because that time goes into checking sources and questioning confidence, the habit that stops a reader from accepting fluent text on tone alone.
What should happen when the AI leaves out a field?
Treat a missing field as information about the answer. Ask for the field again, and if the platform keeps dropping it after an update, reload the custom instructions and rerun the test question. A model that won’t state conflicts or sources is handing you an answer you cannot verify.
Can a well-written system prompt enforce company AI rules?
No prompt can guarantee compliance. A June 2026 session documented on basilpuglisi.com showed the relevant instructions displaced under completion pressure, which is why the article calls a prompt “a request, not a control.” Structure exposes what the model did, while enforcement needs a human checkpoint outside the model that it cannot negotiate away.
Sources
Anthropic. (n.d.). Prompt engineering overview. Claude Platform Docs. https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overview
Puglisi, B. C. (2025). Governing AI: When capability exceeds control (2nd ed.). https://basilpuglisi.com/governing-ai-when-capability-exceeds-control/
Puglisi, B. C. (2025, December 1). AI personalization prompts. basilpuglisi.com. https://basilpuglisi.com/ai-personalization-prompts/
Puglisi, B. C. (2026, March 17). HAIA-RECCLIN: Reasoning and dispatch (3rd ed.). basilpuglisi.com. https://basilpuglisi.com/haia-recclin-reasoning-dispatch/
Puglisi, B. C. (2026, June 12). Why you cannot program or prompt governance into AI. basilpuglisi.com. https://basilpuglisi.com/program-prompt-governance-ai/
Puglisi, B. C. (2026, September 11). What is HAIA-RECCLIN? Reasoning and dispatch. basilpuglisi.com. https://basilpuglisi.com/haia-recclin/
Puglisi Consulting. (n.d.). Advisory reviews. https://puglisiconsulting.com/advisory-reviews/
#AIgenerated from #AIassisted content under HAIA Ecosystem
Free for personal, educational, and noncommercial research use with attribution. Commercial exploitation, paid productization, and enterprise commercialization require separate permission and licensing.
Suggested citation: Puglisi, B. C. (2026). How can I start using structured prompt engineering? basilpuglisi.com.
[…] How can I start using structured prompt engineering? […]