The Google Docs page that saves you 10 hours a month with ChatGPT

If you use ChatGPT (or Claude, or whichever you prefer) several times a week, this has definitely happened to you: you open a new chat, ask for something seemingly simple, get a generic answer back, and have to explain all over again who you are, what you sell, who you sell to, how you speak and what you don't want it to do. Every time. From scratch. After three months of this, you start suspecting the tool doesn't work for your business. But usually it isn't the tool: it's that you're not giving it the one file it needs to stop treating you as a stranger.

We've spent over a year using AI in everything we do at the studio: client emails, commercial proposals, audits, content, data analysis, campaign ideas. And if we had to pick one thing that pushed output quality through the roof, it wouldn't be a special prompt or a chatbot trick: it would be a Google Docs page with seven blocks we keep up to date and paste at the start of every important conversation. Ten minutes of maintenance a month. Hours saved every week. This article is how to write yours this afternoon.

The problem isn't ChatGPT. It's that it starts blank every time you open a chat

When you open a new conversation with any AI, that chat knows nothing about you. It doesn't know what you sell, doesn't know who you sell it to, doesn't know how you write, doesn't know which mistakes you've already made or which decisions you've already taken. Ideally you'd have persistent memory between chats, but in practice that works poorly and every new conversation resets to zero.

So you do one of two things. Either you repeat the whole context at the start each time (which gets tedious after two days, and you end up summarising it badly), or you stop explaining and accept the answers being generic. The second option is what almost everyone picks, which is why many people conclude that AI doesn't work for my business. It isn't true: what doesn't work is starting every chat like it's the first.

The cost is invisible. No error pops up, no alarm. The texts just come back by-the-book, the proposals sound like a template, the emails read like they're from some generic brand, and the ideas are the same ones any other SME anywhere would have. You normalise it. And you start thinking the tool has a ceiling, when really the ceiling is the one you've set by not giving it context.

The page that fixed this at our studio

We ran a test months ago: we created a Google Doc with seven very specific blocks about us (who we are, how we write, which clients we serve, what's already decided, which mistakes we want to avoid, how we want answers, and what we DON'T want). We started short, almost as a draft. We refined it every time the AI gave us something that didn't fit. Within two weeks, that page was the most valuable piece of our AI operation.

The routine is simple. Whenever we open an important chat (drafting a proposal, replying to a delicate email, writing an article, setting up an analysis), we paste the content of that page at the start of the conversation. "This is me, this is my business, this is what I want, and this is how you should respond. Now then: the actual task." Everything changes. The same prompts that used to give flat answers now return replies that sound like us, applied to our reality, respecting from the first line the decisions we've already taken.

It isn't magic. It's that AI can only work with the context you give it, and your Google Docs page is the minimum viable context to stop being a stranger every time.

What each of the 7 blocks should contain

We've tried many structures and this is the one that works for us, adapted to an SME using AI in marketing, sales and operations. The whole page usually runs between 800 and 2,500 words. Beyond that it starts to dilute and the AI stops paying attention to the end of the document.

1. Who you are, in two lines

One sentence about your business, one sentence about who you sell to. Don't put mission, vision or values: that's for the corporate page. Here, what the AI needs is to know where to place you. Example from a rural hotel manager in Ibiza: "I run an 8-room agritourism in inland Ibiza, open year-round, with my partner and a team of 4. Our guests are European couples aged 35 to 55 looking for rest, not parties." Done. Two sentences. It already knows everything it needs.

2. How you write (10 real sentences)

This is the block everyone skips and the one that makes the biggest difference. Don't tell the AI "write in a warm, professional tone": that means nothing. Give it 10 real examples of sentences you've actually written. You can copy them from emails to clients, an Instagram post, replies to reviews, a proposal you sent. Literal. When the AI sees 10 real sentences of yours, it starts imitating you. When it only sees abstract adjectives, it writes like a brochure.

3. Who you sell to and how they tell you

Short description of your typical customer: age, context, what they care about, what they DON'T care about, how they describe things when they get in touch. If you serve villa owners renting on Airbnb who always talk about "bookings", write it down. If your clients are local businesses speaking in the first-person plural ("we have a restaurant"), write it down. This is what turns by-the-book answers into answers that sound like someone who knows your market.

4. What you've already decided and don't want to rethink

Short list of 10-15 fixed decisions. Services you offer. Services you DON'T offer (equally important). Minimum rates. Types of client you work with and which you don't. Payment terms. Typical timelines. Channels you're on and ones you aren't. Without this block, every AI conversation ends up suggesting you launch a new TikTok, offer a service you dropped three years ago, or go after a segment you don't care about. With this block, the AI works within your framework.

5. Mistakes you've seen and don't want to repeat

This block is an anti-slip-up insurance policy. Every time the AI gives you an answer that doesn't fit and you have to correct it, force yourself to add one short line to the file explaining what was missing. In a few weeks, you'll have an operational memory of your own learnings. Real example: "Never recommend replying on Instagram with direct WhatsApp links, because Meta has blocked our messages before. Always use 'DM us'." One line. Blocks a mistake you'd have repeated. Multiply that by 40 lines in 2 months and you've got a file worth its weight in gold.

6. How you want answers

Format preferences. If you like short, direct answers, say so. If you want one option with a clear recommendation rather than five to choose from, say so. If you prefer it to push back when it disagrees rather than always nod, say so. Length, language, level of detail, whether you want lists or prose. This block calibrates how it talks to you and saves a huge amount of editing time afterwards.

7. What you hate it doing

Specific bans. Your bans. A few of ours: don't add disclaimers at the end ("before making decisions, consult a professional"), don't ask if I want you to expand when I've already asked for an expansion, don't use "in summary" or "in conclusion", don't offer me "innovative approaches" when I've asked for a concrete plan. Every specific ban is worth more than ten good intentions.

How to write your page in two hours this afternoon

You don't need to plan or hire anyone. If you spend more than two hours on the first version, you're overdoing it. This is the sequence that works:

Step 1 · Open a blank Google Doc (1 minute)

Name it AI Context – [your business]. Share it with yourself somewhere easy to reach (browser bookmarks, pinned in your Drive, or in Notion if you use that). The format doesn't matter: what matters is having it one click away, always.

Step 2 · Start with the mistakes (25 minutes)

Even though it's block 5 in the structure, starting here has a trick: this is where you have fresh material. Think of the last four weeks using ChatGPT or Claude, and note each time you had to correct a reply. For each one: what you asked, what it gave you, what was missing. No need to write elegantly. Literal. This list is already the heart of your file: real lessons, not theory.

Step 3 · Who you are and how you write (20 minutes)

Write your two identity sentences. Then 10 real sentences of yours copied as-is from emails, posts or client replies. Don't invent or polish them: copy them. The contrast between invented text and real text is huge and the AI notices.

Step 4 · What you've already decided (20 minutes)

List 10 to 15 fixed decisions about your operation. Yes/no services, rates, clients, channels, terms. Everything you've already resolved and don't want to renegotiate in every conversation.

Step 5 · How you want answers and what you hate (20 minutes)

Two short lists. The first: how you like answers (length, format, tone, honesty). The second: the bans. The more specific, the better.

Step 6 · Test it on a real task (35 minutes)

Pick something you had pending this week. It can be writing a difficult email, drafting a post, answering a complaint, putting together a proposal. Open a new ChatGPT or Claude chat. Paste the content of your Google Doc at the start. Then ask for the task. Mentally compare with how it would have turned out without the page. The difference will be obvious. Tune the page with what you learn, and repeat with a second task. By the third one you already have a v1 that works and that you're not letting go.

Mistakes we've seen (and made)

If you know them in advance, you save yourself a lot of frustration. These are the most common slip-ups:

• Writing the page like it's for the website's "about us". "We are a leader in…" is noise. The AI doesn't need promotion, it needs useful data. Write it like you'd explain things to a new employee starting tomorrow who needs to adapt quickly.
• Trying to cover everything. A 15,000-word page isn't better than a 2,000-word one — it's worse. The AI stops paying attention to the end of the document and what matters gets diluted. If you don't care whether it knows something, don't put it in.
• Never updating it. You write it with enthusiasm, use it for two weeks and forget about it for three months. It still works but gets worse and worse, because your operation changes and the page doesn't. The update routine (next section) is non-negotiable.
• Being aspirational rather than real. "We work with the sector's best clients" is useless. "Our clients are SMEs of 5-15 employees with budgets between X and Y who have specific problems of type Z" is gold. The AI doesn't understand aspirations. It understands data.
• Adding sensitive information without thinking. No passwords, bank data, ID numbers, personal client information or trade secrets you wouldn't want to see on Twitter. Assume that anything you paste into an AI chat could, theoretically, end up in a log someone reads. Treat the page as if it were public.

Examples of applying it to an SME's day-to-day

This page changes very specific things in real tasks. A few examples we use daily or have set up with clients:

• Responding to quotes. Paste the doc, add the client's data and brief, and ask for the proposal. It comes out in your tone, with your real rates, without offering services you don't provide, with the terms you always use. 15 minutes for something that used to take an hour.
• Writing emails to difficult clients. Paste the doc, explain the situation (angry client, complaint, misunderstanding) and ask for the draft. The response keeps your tone, respects your policies and avoids the corporate stiffness that grates. Review and send.
• Product or service descriptions. If you sell 50 products and need consistent descriptions across them, pasting the doc prevents each description from sounding like a different brand. Consistent style without rigid templates.
• Social media posts. You generate a month of content in one afternoon, and it doesn't sound machine-written: it sounds like you wrote it in a good mood on a day you had time.
• Seasonal offers. Tourism SME launching a spring offer: paste the doc, describe the offer, ask for copy for email, Instagram, landing page and poster. It comes out consistent across the four without you having to rewrite each one.

What these cases have in common: without the context page, each would have given you a generic answer you'd have had to adapt by hand. With the page, the texts arrive with your DNA already in place and you just review.

How to keep it alive without it becoming yet another pending task

The real risk is writing it with enthusiasm and then leaving it to rot. Two simple routines prevent that:

1. The hot-correction rule. Every time the AI gives you a reply that doesn't fit and you have to correct it, you force yourself to add one line to the file explaining what it would have needed to know. Do it in the moment. If you can't, at least open the page at the end of the day and add what you learned that day in one go. It isn't optional.
2. Monthly 15-minute review. Once a month you open the page, take a look, remove obsolete stuff, rephrase things that no longer sound right, and add what's new. 15 minutes a month to maintain a tool that saves you hours every week. The return is ridiculous.

If these two routines feel like too much, AI isn't playing an important role in your day-to-day, and the context page probably isn't your priority right now. It's useful to know: not every business needs the same things at the same time.

How we do it at Gecko Studio

In the interest of transparency: we maintain several context pages, one per work area. One for client communication (tone, response times, phrasings we use). One for content writing (brand voice, editorial pillars, banned words). One for commercial proposals (rates, services, non-negotiable terms). One for internal operations (fixed procedures, owners, operational decisions). We review them every few weeks and use them as the first context in any AI work that requires judgement.

This is part of what makes our AI outputs sound similar to each other, what makes clients receive consistent pieces, and what stops AI from undoing in 10 minutes what we've spent weeks building in terms of identity and judgement. It isn't glamorous. It won't show up on LinkedIn. It's a document in Google Docs. But it's worth more than any miracle prompt doing the rounds.

If you're building or scaling a project that depends on AI in an important layer and want to start with good judgement from the outset, at Gecko Studio we help design that base. It isn't an off-the-shelf service — we do it bespoke because each business has its nuance. The only important thing: start with the page before the infrastructure. It's cheaper, faster and more effective.

How to start your Google Docs page for ChatGPT this afternoon

If you've read this far and use ChatGPT or Claude frequently in your business, your Google Docs page for ChatGPT is the highest-return piece per hour invested that you can build today. Block off two hours this afternoon, not tomorrow. Open a blank document. In the first hour, list the last 10 mistakes you've had to correct the AI on and write 10 real sentences of yours from recent emails or posts. In the second hour, enumerate 15 fixed decisions about your operation (yes/no services, rates, typical clients, terms) and 10 things you don't want the AI to do (specific bans, not good intentions). That's your v1.

Then, the step that decides whether all this works: take a real task you had pending this week (an email, a proposal, a post) and do it with AI using your new page pasted at the start of the chat. Compare it mentally with how it would have turned out without it. The difference is visible from the first comparison. Keep it alive with two routines: add a line every time you correct something, review for 15 minutes a month. That's it. No more tricks, no magic prompt, no secret tool. Just a Google Docs page that saves you ten hours a month if you write it today.