AI Central

7 Advanced Prompting Techniques

Download
AI Central
Review AI summary

A good prompt still hits a ceiling, and the seven techniques in AI Central's guide, 7 Advanced Prompting Techniques, all break through it the same way: they stop asking the model to do everything in one pass. Examples replace rules, fences separate inputs, a named edit replaces a vague rewrite, and comparisons, chains and critiques give the model the second look it never takes on its own.

Why one long prompt fails

The opening example asks for six jobs in one message: find the causes of Q3 churn, plan the fixes, write the board memo and the customer email. The answer is a comprehensive overview with some general recommendations to consider. AI Central's diagnosis is two lines: one long prompt asks for six jobs in one pass, and it answers once, with no second look.

Everything at once, nothing done well

Pick the technique for the task

  • The format must be exact (labels, tables, scores): show it 2 examples.
  • You paste long material (notes, emails, data): fence it off.
  • A draft needs one change: name the exact edit.
  • A decision with options: compare 3 options first.
  • A big multi-part job: chain prompts, one job each.
  • The answer must get better, or you do not know how to ask: critique and revise, or let the AI write the prompt.

Show it, fence it, name it

The first three techniques fix the inputs. Showing examples means giving the model worked input and output pairs instead of describing the format. The guide's ticket sorter labels two tickets, one easy and one trickier, then ends on a new ticket with the answer left blank. Its own line on why: two examples say more than a paragraph.

Fencing keeps your instructions apart from what you paste. Each pasted text sits under a ### header or a tag with a name (notes, data, template), and the task points to the blocks by name: write this month's recap from Notes, in the shape of Template, under 200 words, with open decisions last.

Naming the edit fixes a common complaint, that asking to rewrite this often gives the same text back. The guide gives four verbs with exact meanings: Simplify (shorter words, same meaning), Shorten (half the length, every fact kept), Neutralize (remove opinion and hype) and Illustrate (add one concrete example). One verb, one clear change.

Compare before deciding

For decisions, AI Central makes the model compare before it answers. State the goal and the limits, ask for 3 options that really differ, score each against the limits, then pick one and say what would change the pick. The example cuts invoice approval from 6 days to 2 under three limits (no new software, 3 approvers, EU audit rules) and scores each option 1 to 5 on speed, cost and audit risk in a table. The final instruction, say what would change your mind, tells you how firm the recommendation is.

Chain the big jobs

Chaining is the direct answer to the churn example. Prompt 1 pulls out the facts, prompt 2 finds the pattern in them, prompt 3 writes the deliverable, and you check each answer before you send the next. The example's first prompt ends with a clean stop: a table of complaints, most frequent first, and nothing else, because the fixes come in the next prompt.

Check each link, fix only the weak one

Critique, then revise

A second pass catches what the first one missed. Give the criteria (reader, goal, length), ask for a score and the 3 weakest points, then ask for a fix of those points only. The example reviews a proposal as a CFO with a $50,000 budget, scores it 1 to 10 on clarity, cost case and risk, and rewrites only the 3 weakest lines, which leaves the parts that already work untouched.

Let the AI write the prompt

When you are not sure how to ask, the last technique has the AI draft the prompt. Describe the task, the reader and the result, ask for a reusable prompt with blanks to fill, then run it in a new chat and keep what works. The example builds a Monday prompt that turns sales call notes into a CRM update, with the weekly parts marked. Use the AI to brief the AI, as the guide puts it.

A target beats a ban

The pro tip swaps each ban for a target. Not don't be wordy but max 3 sentences per paragraph, not no jargon but words a new hire knows. The reason fits in one line:

a ban says what to avoid, a target says where to land

Where to start

Pick one hard task this week and try one technique on it, as the guide suggests. If the task has several parts, chaining is the one to try first, since it fixes the exact failure the guide opens with.