🗺️ Context = map
Bad responses almost always come from poor context
AI doesn’t know your world. It knows the world. When you ask for a campaign, it a campaign — no your campaign. Context is what turns a generic answer into a personalized one.
"Create a campaign for my project."
Result: something that works for any company on the planet. AI wasn’t weak — it was blind.
"Create a campaign for my project." + context folder (audience, tone, good examples, limits)
Result: a campaign that feels like it was made by someone who’s worked there for years.
📦 Types of context
Files, images, history, examples, rules, brand, audience
PDFs, spreadsheets, Word, markdown. You upload them and the AI reads them. Good for: briefs, knowledge bases, contracts (anonymized).
Screenshots, product photos, charts. Claude and GPT-4o read and describe them. Good for: analyzing visual data, describing layouts.
Everything said in the same session is available automatically. Between sessions, you need to summarize manually.
"Write in the style of this text: [example]" or "Avoid this kind of thing: [counterexample]". This is the type of context with the highest return per word.
📎 Attach files and references
How each tool handles what you upload
- ✓PDF, Word, Excel, images, code
- ✓Automatically processes file content
- !Free plans have upload limits
- ✓PDF, images, code, text
- ✓Very large context window (200k tokens)
- ✓Great for long documents
- ✓PDF, images, audio, video
- ✓Google Drive integration
- !Behavior varies by version
For short texts (up to ~2,000 words), pasting works just like uploading—and it's faster. For long documents, uploading is better because it preserves formatting and the AI doesn't need to process raw text.
✅ Positive and negative examples as context
"This is how, this is not" teaches more than instructions
AI learns from patterns. Three paragraphs of instructions about "a casual but professional tone" are equivalent to two real examples of text in the right tone. Examples are concrete patterns — AI replicates them with impressive accuracy.
"Write in an informal but not vulgar tone, with energy, no clichés, suitable for young adults aged 25-35 who work in tech..."
Result: AI interprets "informal" and "energy" its own way.
"Write it like this: [good example]. Avoid this: [bad example]. Now do it for my situation..."
Result: an unambiguous, calibrated tone.
Exemplo BOM (use como referência de tom): "[cole aqui um texto que você aprova]" Exemplo RUIM (evite este estilo): "[cole aqui um texto que você quer evitar]" Agora crie [o que você quer] seguindo o tom do exemplo bom e evitando o estilo do exemplo ruim.
🎨 Brand identity and voice as reusable context
The context you build once and use all the time
Without voice context, every response sounds like it came from a different company. With it, even someone who just opened the tool can produce something in the right tone—because the context carries the consistency standard.
E.g.: "direct, empathetic, technical without jargon." Simple and concrete — AI knows what to do with that.
E.g.: "never use the word 'synergy,' don't promise financial results, avoid humor when the topic is a customer's problems."
Copy an excerpt from a text of yours that you consider to have the right tone. That's the most powerful style anchor.
Save it as an "identity block" somewhere easy to access. 30 seconds of copy and paste ensures consistency throughout the session.
⚖️ When too much context gets in the way
Noise × signal: what to include and what to cut
- →Who is the audience for the result
- →What is the specific goal of this task
- →Examples of the right tone
- →Constraints that affect this result
- →Decisions made previously
- →The company's history from the last 10 years
- →Details of unrelated projects
- →Reasons why you want this
- →Information AI already has by default
- →All your documents "just in case"
Before adding a context item, ask: "If I remove this, will the answer get significantly worse?" If not, leave it out. Context has a cost — it dilutes attention, increases tokens, and sometimes confuses the AI with contradictions.
🔒 Privacy and data: what to NEVER paste into AI
Would you paste this in a public group?
"Would you post this in a public WhatsApp group?" If not — don't post it to AI either.
AI tools have different policies on data use. Some use conversations for training. Sensitive data pasted by accident could be exposed in breaches or violate Brazil’s LGPD data protection law.
- ✗Identifiable client data (name, CPF, email, phone)
- ✗Passwords, API tokens, access keys
- ✗Confidential contracts without anonymization
- ✗Patient health data
- ✗Real customers’ financial information
- ✗Trade secrets and proprietary IP
- →Replace names with [NAME] or [CLIENT A]
- →Companies: [COMPANY] or [SUPPLIER]
- →CPF/CNPJ: [DOCUMENT]
- →Values: keep the order of magnitude, change the number
- →Health data: describe the condition without identifying the patient
[ ] Sem nomes reais de clientes/pacientes/funcionários [ ] Sem CPF, CNPJ, RG, passaporte [ ] Sem senhas, tokens ou chaves de API [ ] Contratos: anonimizados com [EMPRESA] e [PARTE] [ ] Dados financeiros: valores genéricos, não reais [ ] Saúde: condição descrita, paciente não identificado [ ] Segredos industriais: estrutura descrita, dados removidos [ ] Regra de ouro: "passaria no grupo público?" ✓
📋 Module 3.1 Summary
What you learned
- ✓Context is the map — poor responses come from poor context
- ✓The 7 types of context you can provide
- ✓How each tool processes uploads
- ✓Positive/negative examples as context
- ✓Brand identity as reusable context
- ✓When to cut context (noise × signal)
- ✓What to NEVER paste + how to anonymize
Connection to the pyramid
T1 Prompt — taught you to structure the request. But a structured request about an unfamiliar world still produces a generic result.
T2 Intent — brought clarity to the goal. A clear goal without context about who is doing the work and for whom remains up in the air.
T3 Context (here) — provides the map. Prompt + intention + context = a result that feels like it came from someone on the inside.
Context folder + privacy checklist
Choose a real project. Anonymize a document. Set up the first files in your context folder. Apply the privacy checklist before using any sensitive data.