How I Use AI Tools Responsibly at Work
Most AI Advice Misses the Point
Direct answer: AI tools aren't interchangeable. Each one has a specific strength, a set of limitations, and a context where it works best. The real skill isn't "using AI." It's knowing which tool fits which task, where the risks are, and what still requires human judgment.
Most AI advice falls into one of three categories: breathless enthusiasm, vague warnings, or shallow listicles. None of those help you actually do better work.
The conversation I keep having with founders, marketers, and operators sounds the same: "I know I should be using AI, but I don't know where to start, which tools to trust, or how to avoid making expensive mistakes."
That's a reasonable position. The problem isn't a lack of tools. The problem is a lack of judgment about tools.
This article is how I actually use AI in my work at Veriqo Studio. Not aspirationally. Not theoretically. The actual stack, the actual rules, the actual failure modes I have learned to watch for.
What this article covers
- Which tool fits which type of work - What each tool is genuinely good at - Where the limitations and failure modes are - What still requires human judgment - How to use AI without creating privacy, quality, brand, or compliance risk - Real prompts I would actually use - A decision framework for choosing the right tool
AI Tools Are Not Interchangeable
This is the single most important point in the entire article.
Treating all AI tools the same way produces:
- Generic output. When you use a research tool for drafting or a drafting tool for research, the result is mediocre at both. - Bad automation. Automating a process that isn't well-defined creates faster chaos, not faster results. - Poor sourcing. Relying on a tool that doesn't cite sources for research-grade work is a credibility risk. - Overconfidence. Polished output from any AI tool can mask factual errors, weak reasoning, or missing context. - Hidden risk. Using the wrong tool in the wrong context can create privacy violations, IP exposure, or brand damage you don't notice until it's too late.
The rest of this article is organized around that principle: each tool has a job, and using the right tool for the right job is the whole game.
The Tool-by-Tool Breakdown
Make.com. Process Orchestration
Best for: Repeatable workflows, app-to-app automation, process orchestration, operational handoffs, alerts, routing, enrichment, and structured automations.
What I personally use it for: Connecting tools that don't natively integrate. Building intake-to-delivery pipelines. Automating lead routing, status updates, and notification workflows. Creating structured handoffs between systems so nothing falls through the cracks.
Where I skip it: Anything that requires judgment, strategy, or nuanced decision-making. Make.com executes logic. It doesn't evaluate whether the logic is right.
Known limitations: Workflows can become brittle if the underlying process isn't well-documented. Error handling requires explicit design. Complex branching creates maintenance burden. If the process changes frequently, the automation breaks frequently.
Ethical rule: If the process is fragile, undocumented, or high-risk, fix the process before you automate it. Automating a bad process creates faster mistakes.
Example prompts I would actually use:
- "Map this marketing intake process into a deterministic workflow with triggers, branches, error handling, and exception paths." - "Design a Make.com scenario that routes new form submissions to the correct team member based on service type, with fallback handling and confirmation notifications." - "Audit this existing automation for single points of failure, missing error states, and undocumented assumptions."
Notion. Knowledge Organization
Best for: Organizing context, documenting systems, connecting messy thinking, turning scattered notes into usable knowledge, project memory, and searchable decisions.
What I personally use it for: Internal documentation, project wikis, decision logs, meeting note synthesis, and building searchable knowledge bases for client engagements.
What it's not for: As a perfect source of truth without maintenance. Notion AI is useful, but it's only as good as the underlying workspace quality. If your documentation is messy, outdated, or incomplete, AI summaries of that documentation will be confidently wrong.
Known limitations: Garbage in, garbage out still applies. Notion AI can summarize content that's outdated or contradictory without flagging the problem. It works best when the underlying workspace is well-organized and actively maintained.
Ethical rule: Do not treat AI-generated summaries from Notion as verified facts. They are synthesis of whatever is in the workspace, including outdated or incorrect information.
Example prompts I would actually use:
- "Review these meeting notes, decisions, and project docs and turn them into an updated source-of-truth page with action items, owners, and open questions." - "Summarize the last four weeks of project updates and flag any items that were assigned but never completed." - "Find all references to the Q2 campaign plan across this workspace and identify any conflicting information."
ChatGPT. Drafting and Analysis
Best for: Drafting, structuring ideas, refining writing, pressure-testing arguments, analysis, turning rough material into something usable, and working with uploaded files and data.
What I personally use it for: First-pass drafts, restructuring messy documents, audience-specific messaging variations, analyzing uploaded spreadsheets and reports, and pressure-testing strategic arguments.
Off-limits uses: Final authority on facts, research source without verification, or as a safe dumping ground for confidential data.
Known limitations: ChatGPT is fast and versatile, but polished doesn't equal correct. It can generate confident, well-structured output that contains factual errors, unsupported claims, or subtle logical gaps. It's a very strong working partner for first-pass work, but it still requires human oversight.
Ethical rule: Use privacy-aware settings. Do not upload sensitive client data without understanding the data handling implications. Always verify claims, citations, and statistics independently.
Example prompts I would actually use:
- "Take these rough notes and turn them into three audience-specific messaging versions, then challenge the assumptions and identify what still needs evidence." - "Analyze this spreadsheet of campaign results and identify the three most actionable patterns, with caveats about sample size and attribution." - "Rewrite this service description for a technical audience, then a non-technical audience, and flag where the core value proposition changes between versions."
Canva. Visual Production
Best for: Fast visual communication, social graphics, presentations, lead magnets, reusable brand templates, and accelerating production.
What I personally use it for: LinkedIn graphics, presentation decks, PDF resources, social media templates, and quick visual concepts for client deliverables.
Where it falls short for me: Replacing foundational brand thinking, skipping brand governance, or generating visuals without reviewing typography, hierarchy, accessibility, and factual claims.
Known limitations: Canva's AI features can produce attractive output that drifts from your brand guidelines. Templates can create visual consistency at the cost of looking generic. The speed advantage only works when paired with clear brand rules.
Ethical rule: Review every AI-generated visual for brand alignment, accessibility, accuracy of any text content, and appropriateness before publishing. Fast doesn't mean finished.
Example prompts I would actually use:
- "Create a clean LinkedIn carousel concept for this article using Veriqo Studio's tone: editorial, smart, anti-hype, minimal, credible." - "Design a one-page PDF checklist based on this content, with clear hierarchy, scannable sections, and a professional layout." - "Generate three social graphic variations for this blog post that emphasize different key takeaways."
Perplexity. Research and Current Awareness
Best for: Fast current-awareness research, initial research paths, source-backed summaries, and identifying what to verify next.
What I personally use it for: Getting oriented on a topic quickly, finding recent developments with citations, market scans, competitive research starting points, and current topic exploration.
Not the right fit for: End of research, substitute for reading key original sources, or guarantee of source quality.
Known limitations: Perplexity provides citations, but citations don't equal rigor. The sources it surfaces may be incomplete, outdated, or low-quality. It's a starting point, not a finish line.
Ethical rule: Always assess source quality, recency, and completeness independently. Use Perplexity to find what to read, not to replace reading.
Example prompts I would actually use:
- "Give me a current overview of how B2B marketers are using AI agents in operations, grouped by use case, with citations and original sources only." - "What are the most significant changes to Google's search quality guidelines in the last 6 months, with links to primary documentation?" - "Find recent case studies of companies implementing fractional marketing leadership, with a focus on outcomes and measurement."
Claude. Synthesis and Long-Form Reasoning
Best for: Nuance, synthesis, long-form writing, thoughtful reframing, complex reasoning, and artifact-style content.
What I personally use it for: Sensemaking work, strategic document development, long-form writing that requires depth and polish, reframing complex topics for different audiences, and building structured artifacts like frameworks and guides.
Where I draw the line: Skipping editorial review, treating output as factually verified by default, or replacing human judgment on sensitive topics.
Known limitations: Claude produces elegant, well-reasoned output, but elegant language can still contain errors, unsupported assumptions, or blind spots. The quality of the reasoning is strong, but verification is still your responsibility.
Ethical rule: Do not let the quality of the writing substitute for the quality of the thinking. Review all outputs for accuracy, completeness, and appropriateness before using them.
Example prompts I would actually use:
- "Take this messy strategic memo and rewrite it into a board-ready brief with clearer logic, stronger sequencing, and explicit assumptions." - "Analyze this marketing plan and identify the three weakest assumptions, then suggest how to test each one with minimal investment." - "Reframe this technical product description for a non-technical buyer audience, preserving accuracy while improving clarity and reducing jargon."
Gemini. Google Ecosystem Work
Best for: Google-centric workflows, research, brainstorming, synthesis across Workspace context, and docs/files/email-adjacent work.
What I personally use it for: Work that lives inside Google's ecosystem. document synthesis, email drafting, research that connects to existing Workspace files, and brainstorming within the context of ongoing projects.
Boundaries I set: Automatically trusting output just because it's integrated into a familiar environment. Integration convenience doesn't equal accuracy.
Known limitations: The ecosystem advantage is real, but it can create false confidence. Because Gemini works within your existing context, it can feel more authoritative than it is. The same source judgment and privacy discipline applies.
Ethical rule: Apply the same verification standards you would apply to any other AI tool. Convenience doesn't change the accuracy requirements.
Example prompts I would actually use:
- "Review these Docs, notes, and product files and produce a concise launch brief with dependencies, risks, owners, and unanswered questions." - "Summarize the key decisions from the last month of meeting notes and flag any action items that are still unresolved." - "Draft a project status update based on these documents, organized by workstream, with explicit callouts for blocked items."
My Ethical Rules for Using AI at Work
These aren't aspirational guidelines. These are rules I follow in every engagement.
- I don't put sensitive or confidential information into tools casually. Every tool has different data handling policies. I know what goes where, and I default to caution. - I verify facts, claims, and citations. AI-generated content can be confidently wrong. I check. - I don't let polished output replace judgment. Well-written isn't the same as well-reasoned. I review for substance, not just style. - I don't automate unstable processes before fixing them. Automation amplifies whatever it touches, including dysfunction. - I don't use AI to create fake authority or pretend expertise. If I don't know something, I say so. AI doesn't change that standard. - I don't use AI-generated visuals or text without reviewing for bias, accuracy, accessibility, IP risk, and brand fit. Every output gets a human review pass. - I use AI to reduce friction, not remove accountability. The tools do work. I own the results.
What Each Tool Gets Wrong
Every tool has failure modes. Knowing them is more valuable than knowing the features.
- Make.com: Brittle workflows that break when upstream tools change. Hidden logic that only the person who built it understands. Bad exception handling that fails silently instead of failing loudly. - Notion: False confidence from messy documentation. AI summaries that synthesize outdated or contradictory information without flagging the problem. Knowledge bases that look organized but contain stale data. - ChatGPT: Polished overconfidence. Output that reads well but contains factual errors, unsupported claims, or circular reasoning. Weak sourcing when used lazily for research. - Canva: Pretty but generic design that drifts from brand guidelines. Weak visual hierarchy that prioritizes aesthetics over communication. AI-generated elements that create accessibility or accuracy issues. - Perplexity: Citations that create an illusion of rigor. Sources that are incomplete, outdated, or low-quality. Speed that discourages the deeper reading that research actually requires. - Claude: Elegant language that can still contain errors, unsupported assumptions, or gaps in reasoning. Output that's so well-written it discourages critical review. - Gemini: Convenient context that can produce flawed synthesis. Integration that creates false confidence because the output feels native to your workflow.
How I Decide Which Tool to Use
Before I open any tool, I ask five questions:
1. Is this task repeatable or judgment-heavy? If it's repeatable and well-defined, consider automation (Make.com). If it requires judgment, use a thinking tool (Claude, ChatGPT).
2. Do I need sourcing? If I need current information with citations, start with Perplexity. If I need to analyze existing data, use ChatGPT or Gemini.
3. Do I need long-form reasoning? If the task requires depth, nuance, or complex synthesis, Claude is usually the strongest choice. For faster iteration, ChatGPT works well.
4. Does this work live inside a certain ecosystem already? If it's Google Workspace-native, Gemini has a contextual advantage. If it's in Notion, use Notion AI. If it needs visual output, use Canva.
5. What is the privacy and compliance risk? If the content is sensitive, confidential, or client-specific, choose the tool with the strongest privacy controls, or don't use AI at all.
Quick decision matrix
- Automate a repeatable workflow → Make.com - Organize and synthesize internal knowledge → Notion - Draft, analyze, or restructure content → ChatGPT - Create visual assets quickly → Canva - Research with citations → Perplexity - Deep reasoning or long-form writing → Claude - Google Workspace-integrated work → Gemini - Sensitive or confidential content → Human judgment first, tool second
What This Looks Like in Real Work
Workflow 1: Building a Responsible Content Pipeline
1. Perplexity: Research the topic. Gather current sources, identify key themes, and note what the existing conversation is missing. 2. Claude: Synthesize the research into an outline with a clear argument structure, explicit assumptions, and identified gaps. 3. ChatGPT: Draft the full article from the outline, iterating on tone, audience fit, and messaging clarity. 4. Human review: Verify all claims, check sources, review for brand alignment, and edit for voice. 5. Canva: Create supporting visuals. social graphics, pull-quote cards, or a PDF version.
Every step has a human checkpoint. No step is fully delegated.
Workflow 2: Turning Scattered Research into a Structured Brief
1. Notion: Gather all existing notes, meeting transcripts, and project documents into a single workspace. 2. Notion AI: Generate an initial summary of the collected material, flagging gaps and contradictions. 3. Claude: Take the summary and restructure it into a decision-ready brief with clear recommendations, risks, and next steps. 4. Human review: Validate the brief against the original sources, add context that only a person with project knowledge would have, and finalize.
Workflow 3: Automating Marketing Admin Without Automating Judgment
1. Human: Document the process end-to-end, including decision points, exceptions, and approval requirements. 2. Make.com: Build the automation for the repeatable, deterministic parts. routing, notifications, data movement, status updates. 3. Human: Keep judgment-heavy steps manual. approval decisions, quality review, client communication. 4. ChatGPT: Generate templates and documentation for the manual steps so they are consistent and trainable.
The principle is the same every time: automate the predictable parts, keep humans on the judgment-heavy parts, and document everything.
Frequently Asked Questions
What is the best AI tool for marketing?
There is no single best tool. The right tool depends on the task. ChatGPT is strong for drafting and analysis. Claude excels at long-form reasoning. Perplexity is best for research with citations. Make.com handles automation. The best approach is matching the tool to the specific type of work.
Which AI tool is best for research?
Perplexity is the strongest starting point for research because it provides citations and source links. However, it's a starting point, not a replacement for reading primary sources. For analyzing existing data and documents, ChatGPT and Gemini are strong alternatives.
What is the difference between ChatGPT, Claude, Gemini, and Perplexity?
ChatGPT is a versatile generalist. strong at drafting, analysis, and working with uploaded files. Claude specializes in nuanced reasoning and long-form synthesis. Gemini integrates with Google Workspace for context-aware work. Perplexity focuses on research with source citations. They are complementary, not interchangeable.
When should you use automation tools like Make.com?
Use Make.com when a process is repeatable, well-documented, and deterministic. Good candidates include data routing, notifications, status updates, and app-to-app integrations. Do not automate processes that require judgment, are poorly documented, or change frequently.
How do you use AI responsibly at work?
Start with clear use cases and defined boundaries. Verify all outputs. Do not upload sensitive data without understanding the privacy implications. Do not let polished output substitute for critical thinking. Use AI to reduce friction, not remove accountability.
What should you avoid putting into AI tools?
Avoid entering confidential client data, proprietary business information, personal identifiable information (PII), trade secrets, and any content subject to legal privilege. When in doubt, don't enter it. Check each tool's data handling and privacy policies before using it for sensitive work.
Working With AI, Not for AI
The tools described in this article are genuinely useful. They save time, reduce friction, and help produce better work when used with judgment.
But they are tools. They don't replace the thinking, the standards, the accountability, or the judgment that makes work trustworthy.
If you are building AI into your marketing operations, content workflows, or business systems, the goal should not be "use more AI." The goal should be: use the right tool, for the right task, with the right guardrails, and keep a human accountable for the result.
That's what responsible AI adoption looks like in practice.