The short version: If ChatGPT, Claude, or Gemini keeps giving you mediocre answers, the model isn’t the problem. Your context is. Here’s how to use the META framework to actually get useful output, with Houston-flavored examples you can copy and paste today.
The real problem: it’s not the AI, it’s the prompt
Most people throw one-liners at AI like “summarize this” or “write me an email” and then complain the output is generic. Of course it is. You gave it nothing to work with.
What matters isn’t what you ask the model, but the context you give it to work in.
A good prompt is just a trigger. Context engineering is the whole system that guides the model with precision. That’s where META comes in.
What is the META framework?
META is four building blocks you give the model before asking it to do anything serious:
- M – Model: Who should the assistant be? (role, experience, tone)
- E – Environment: Why are you doing this and for whom? (goal, audience, impact)
- T – Task: What facts, rules, and constraints matter?
- A – Adjustments: How should it work and deliver? (steps, format, review points)
This works the same with ChatGPT, Claude, Gemini, Copilot, or Perplexity. The model doesn’t change — the quality of your input does.
The base template (copy this)
MODEL: Act as [role + experience + approach]. ENVIRONMENT: I need [goal]. The output will be used by [audience]. Expected impact: [desired outcome]. TASK: 1. [Key data, source, or document] 2. [Constraint or rule] 3. [Previous example if applicable] ADJUSTMENTS: 1. First, tell me if you're missing any information (gap check). 2. Propose a structure. 3. Draft a first version. 4. Wait for my feedback before polishing. Final format: [format]. First action: start with the gap check.
Real examples (the kind that actually work)
1. Counter-offer email on a home in The Heights
MODEL: Act as a Houston-area real estate buyer's agent with 15 years of experience negotiating in hot neighborhoods. Tone: firm but professional, no aggressive language. ENVIRONMENT: I need a counter-offer email to the listing agent on a property in The Heights (77008). The seller is at $625K, we're offering $590K. The seller's agent will forward this to the owner. Goal: close the gap without losing the deal to a backup offer. TASK: - Property has been listed 38 days (above the 21-day Heights average). - HCAD appraised value is $548K. - Comps in the same block sold $575K-$605K in the last 90 days. - Foundation report shows minor settling, $8K repair estimate. - We're cash buyers, can close in 14 days. ADJUSTMENTS: 1. Subject line plus body, under 200 words. 2. Lead with our strengths (cash, fast close), then justify the price. 3. Don't sound desperate or threaten to walk. 4. End with a specific response deadline (48 hours).
2. Incident report for a refinery near-miss
MODEL: Act as an HSE manager with experience in Gulf Coast refineries and OSHA Process Safety Management (PSM) standards. ENVIRONMENT: I need a near-miss incident report for a contractor who almost contacted an energized 480V panel during maintenance in our Pasadena facility. The report goes to the Plant Manager and gets filed for OSHA recordkeeping. Goal: clear documentation that drives corrective action without sounding like blame. TASK: - Date/time: included in attached notes. - Contractor was working under a Hot Work Permit, no LOTO on the panel. - Discovered during a routine safety walk-down. - No injury, no equipment damage. - Reference: 29 CFR 1910.147 (LOTO) and our internal SOP-EHS-104. ADJUSTMENTS: 1. Standard format: Summary, Sequence of Events, Root Cause (5 Whys), Corrective Actions, Responsible Parties. 2. Stick to facts. No speculation. 3. Corrective actions must be SMART (specific, measurable, etc.). 4. One page max.
3. LinkedIn post for a Houston founder
MODEL: Act as a B2B LinkedIn copywriter who's written for Texas-based SaaS and energy-tech founders. ENVIRONMENT: I need a LinkedIn post for my personal profile. Audience: energy company operations leads and VPs in Houston. Goal: position me as a practical voice on industrial AI and generate 3-5 sales conversations. TASK: - Topic: how we cut a midstream client's pipeline inspection reporting time from 6 hours per report to 25 minutes using AI summarization. - Numbers: 240 reports per month, ~1,400 hours saved annually. - No buzzwords (skip "game-changer," "revolutionary," "disrupting"). - Plain English, no emojis except maybe one at the end. ADJUSTMENTS: 1. Hook in the first line — something that makes someone stop scrolling. 2. Structure: problem → what we did → result → invitation to discuss. 3. Max 1,200 characters. 4. Soft CTA at the end (invite a comment, don't pitch).
4. Hurricane prep memo to staff
MODEL: Act as a business continuity manager with experience running Gulf Coast operations through major storms (Harvey, Beryl). ENVIRONMENT: I need an internal memo to all employees of our Spring, TX office (45 people) ahead of a Category 2 forecast making landfall in 72 hours. The memo goes out from the COO. Goal: clear actions, calm tone, no panic. TASK: - Office closes at 3 PM tomorrow, reopens when safe. - Remote work expected Wednesday through Friday at minimum. - Critical systems: payroll deadline Friday, can't slip. - Power outages likely; CenterPoint outage map link required. - Employees in flood-prone zips (77024, 77079, 77450) get priority on remote setup. ADJUSTMENTS: 1. Structure: situation → office status → what you need to do → who to contact. 2. Bullet-heavy, no long paragraphs (people skim during prep). 3. End with one emergency contact number. 4. Under 400 words.
5. Social media plan for a Montrose restaurant
MODEL: Act as a social media manager experienced with independent restaurants in Houston's inner loop. ENVIRONMENT: I need a 30-day Instagram plan for a new Tex-Mex spot in Montrose. Audience: Houstonians 25-40 who follow food accounts like @houfoodie and @eaterhouston. Goal: drive weekend brunch reservations up 30%. TASK: - Zero production budget. All content shot on iPhone in the kitchen. - 4 feed posts per week plus daily Reels/Stories. - Differentiator: third-generation family recipes from Brownsville. - Don't lean on "best taco in Houston" — too saturated. - We can offer one collab dinner per month for influencer outreach. ADJUSTMENTS: 1. Week-by-week calendar in a table. 2. For each post: format, suggested caption, base hashtags. 3. Include 5 Reel ideas that need zero editing skills. 4. Flag 3 local accounts to engage with weekly.
6. Vendor RFP response for City of Houston
MODEL: Act as a senior proposals writer with experience winning municipal contracts in Texas (City of Houston, Harris County, METRO). ENVIRONMENT: I need to draft the "Approach and Methodology" section of an RFP response for the City of Houston Public Works department. The evaluation panel scores this section out of 40 points and we need at least 32 to be competitive. TASK: - Scope: GIS mapping services for stormwater infrastructure, 18-month contract. - RFP document attached (PDF). - Must align with the City's MWBE goals (at least 17% subcontracting). - Reference our prior work with TxDOT and Harris County Flood Control. - Don't promise deliverables outside the stated scope. ADJUSTMENTS: 1. First, tell me which RFP requirements I haven't addressed yet. 2. Propose a section outline (use the RFP's numbering). 3. Draft section by section, wait for my approval between each. 4. Format: Word document, numbered 4.1, 4.1.1, etc.
When NOT to use META
METAing every prompt is overkill. Use a direct prompt when:
- You need to convert a format: “turn this into a table”
- You want a translation or grammar fix: “rewrite this in a professional tone”
- It’s a simple calculation: “what’s 8.25% sales tax on $4,280?”
- You need a one or two-sentence answer
- It’s quick classification: “is this customer review positive or negative?”
Rule of thumb: if the task has multiple steps, will be seen by a client or executive, or depends on specific sources, use META. Otherwise, fire off a direct prompt.
Common mistakes (and how to dodge them)
- Dumping context without filtering: pasting nine pages of contract clauses without flagging what’s reference, what’s base material, and what’s a hard constraint. The output will be just as muddled.
- Contradictory instructions: “be creative but follow this exact structure.” Pick one.
- Skipping the gap check: letting the model draft before confirming it understood. You’ll burn 20 minutes on rewrites.
- Copy-pasting templates without adapting: META is a framework, not a fill-in-the-blank.
The trick that fixes 40% of bad outputs
Before kicking off a long prompt, add this at the end:
“Before you start, summarize in one sentence what you understand I need, and tell me if anything is ambiguous or missing.”
Two seconds extra, and you save yourself a round of corrections. Works in any model.
Your plan for this week
- Pick one task you do regularly with AI (emails, summaries, posts — whatever).
- Convert it to META format on a simple template.
- Use it as your base prompt three times this week.
- Compare: fewer corrections? Cleaner output? If yes, you’ve got your first reusable template. If not, find which block was weak and fix it.
You don’t need to be a developer. You don’t need expensive tools. You don’t need the latest model. You need clarity on what you want, structure for expressing it, and a framework you can improve. That’s META.
If this was useful, share it with someone on your team who’s still throwing one-liners at ChatGPT and hoping for magic. You’ll save them hours.