How to use ChatGPT effectively in 2026 [Tips + Prompts]

Vague prompts produce vague output. That’s the reason most people using ChatGPT daily still get mediocre results.

Most people use it like a search engine — type a question, read the answer, move on. That approach saves you maybe five minutes and explains why so many people think the tool is overrated. 

The problem isn’t ChatGPT. It’s that no context means no relevance, and default settings leave the most useful features untouched.

At Expandi, we work with thousands of B2B sales teams and lead gen agencies running LinkedIn outreach at scale. The pattern we see constantly: people using ChatGPT every day and still getting output they have to heavily rewrite or throw away entirely. 

This guide covers what they’re doing wrong and how to fix it – from prompting frameworks to the 2026 features most people haven’t touched. Here’s what we’ll cover:

• How ChatGPT works and why many people are using it wrong.
• How to choose the right AI plan and model.
• How to use ChatGPT with better prompting.
• Tips for using ChatGPT for writing.
• How to use ChatGPT for B2B sales and LinkedIn outreach.
• ChatGPT features worth using in 2026.
• Common ChatGPT mistakes to avoid.

Key Takeaways

• ChatGPT output quality is almost entirely determined by prompt quality. Vague prompts produce vague results.
• Use the PTCF framework (Persona, Task, Context, Format) as a baseline if you want to use ChatGPT effectively.
• Always verify specific claims, stats, and quotes that ChatGPT generates before using them. It hallucinates with confidence.
• For B2B outreach, use ChatGPT to generate and iterate on copy, then pair it with a dedicated outreach tool like Expandi to run campaigns at scale.

How ChatGPT works (and why most people use it wrong)

ChatGPT is an AI chatbot developed by OpenAI that generates text, images, and structured output based on the instructions you give it. It is trained on large volumes of text data and uses that training to predict useful, contextually relevant responses.

The platform now has over 900 million weekly active users as of February 2026. According to OpenAI’s State of Enterprise AI report, workers using it effectively save 40-60 minutes per active workday. The gap between that and ‘five minutes saved’ is entirely down to how you prompt it.

The key mechanic to understand is that ChatGPT does not look up answers. It generates them based on patterns in its training and when web search is enabled, real-time sources. That distinction matters because the quality of your input directly shapes the quality of the output.

A vague prompt gets a vague answer. A specific, well-structured prompt with context and constraints gets something you can actually use.

How to choose the right ChatGPT plan and model

The free plan is fine for occasional use and quick tasks. For anything work-related – outreach copy, research, iterative drafts – the limitations on message volume, model access, and features will slow you down enough to matter.

Here is what each plan gets you, based on official ChatGPT pricing:

Plan	Price	Best for
Free	$0/month	Occasional use. Limited GPT-5.3 access, limited messages, limited deep research, limited image generation.
Go	$8/month	More capacity than Free. More messages, uploads, image creation, and longer memory. Note: may include ads.
Plus	$20/month	Daily users. Includes GPT-5.4 Thinking, expanded messages and uploads, deep research, agent mode, tasks, Codex, Sora, projects, and custom GPTs.
Pro	$200/month	Power users needing maximum access. GPT-5.4 Pro, unlimited GPT-5.4, maximum deep research and agent mode, expanded Codex and Sora. ChatGPT Pulse included.
Business	$25/user/month (annual)	Teams of 2+ needing a dedicated workspace. Unlimited GPT-5.4 messages, SAML SSO, admin console, SOC 2 Type 2, no training on business data. 60+ app integrations.
Enterprise	Custom (contact sales)	Large orgs needing SCIM, Enterprise Key Management, data residency in 10 regions, role-based access, analytics dashboard, dedicated onboarding, and 24/7 priority support.

Note on the Go plan: At $8/month it gives you more capacity than Free but may include ads in the US. For regular work use, Plus at $20/month is the better call. Pricing and model access based on official OpenAI pricing page as of writing (March 2026). OpenAI updates plans frequently, verify current details on the official page.

Which ChatGPT model should you use?

ChatGPT selects a model automatically based on your prompt and plan. On Plus and above, you can choose manually. The right choice depends on the task:

Model	Available on	Use it for
GPT-5.3 Instant	All paid plans	Everyday tasks: emails, quick drafts, Q&A, summaries. Fast and conversational.
GPT-5.4 Thinking	Plus, Business, Pro, Enterprise	Complex reasoning, strategy, analysis, multi-step prompts. Shows thinking plan upfront so you can redirect mid-response.
GPT-5.4 Pro	Pro and Enterprise only	Maximum performance. Harder problems, professional knowledge work, long-horizon tasks like financial models or legal analysis.
o3 / o4-mini	Plus and above (via API)	Logic-heavy tasks, coding, structured data analysis. More cost-efficient for high-volume reasoning.

Note: GPT-5.2 Thinking is being retired on June 5, 2026. All Plus, Business, Pro, and Enterprise users have been migrated to GPT-5.4 Thinking. If you have workflows built around GPT-5.2 behaviour, test them against GPT-5.4 now.

A useful way to think about it, per OpenAI’s own documentation: 

• GPT-5.4 Thinking is like a senior co-worker – give it a goal and trust it to work out the details. 
• GPT-5.3 Instant is like a junior co-worker – it performs best with explicit, specific instructions. 

That distinction should guide which model you reach for and how you write the prompt.

How to use ChatGPT effectively with better prompting

Prompt quality is the single biggest factor in whether ChatGPT is useful or frustrating. Most people treat it like a Google search – short, keyword-driven, no context. The frameworks below consistently produce better output.

Give ChatGPT a role first

Before stating your task, assign a role. This primes the model to respond from a specific perspective and improves relevance and tone.

Example:

Instead of a generic “Write a LinkedIn connection request” prompt.

Try:

“You are a B2B sales expert writing outreach for a LinkedIn automation platform. Your audience is heads of sales at 20-50 person SaaS companies. Write a connection request that is direct, specific, and under 300 characters.”

The role does not need to be elaborate. Even a one-line persona – “You are a direct-response copywriter” or “You are a senior B2B marketer” – shifts the register and focus of the output.

Use the PTCF framework

PTCF stands for Persona, Task, Context, Format. It removes ambiguity from your prompts by giving ChatGPT a clear structure to follow.

• Persona: Who should ChatGPT be? (“You are a senior SDR at a LinkedIn automation platform…”
• Task: What specifically do you want it to do? (“Write a 3-message follow-up sequence for a prospect who went cold after showing interest…”
• Context: What background does it need? (“The prospect is a Head of Sales at a 30-person SaaS company. They booked a demo but didn’t show up…”)
• Format: How should the output be structured? (“Return each message as a numbered block with a subject line and body under 100 words. No bullet points…”)

Most bad prompts are missing two or three of these. Context is the most commonly skipped variable that separates generic LinkedIn outreach from something worth replying to

Add constraints, not just instructions

Telling ChatGPT what not to do is as important as telling it what to do. Constraints prevent generic, padded, or off-tone output.

Constraint examples:

• “Do not mention our product in the first message.”
• “Keep the total response under 150 words.”
• “Avoid corporate language and filler phrases.”
• “Do not start with a compliment or pleasantry.”
• “Do not suggest scheduling a call in the first message.”

This is especially useful for outreach copy, where default output tends toward over-formal, generic text that sounds like every other cold message in someone’s inbox.

Prompt in layers, not all at once

Complex tasks produce better results when broken into steps. Start broad, then refine – rather than trying to get perfect output from a single prompt.

• Step 1: “Give me an outline for a 5-step LinkedIn outreach sequence targeting SaaS founders.”
• Step 2: “Now write step 1 as a connection request. Keep it under 280 characters, casual tone.”
• Step 3: “The prospect accepted. Write follow-up message 1. Reference that we have not spoken before and ask one specific question.”

Layered prompting also makes it easier to catch and correct problems early rather than asking for a full rewrite at the end.

Use the ‘Poke Holes’ method

Feed ChatGPT a cold email or connection request sequence and ask it to find the weaknesses. This works for any outreach asset: follow-up sequences, InMail templates, LinkedIn profiles, pitch decks, or even LinkedIn content. Anything you want to stress-test before it goes live

Prompt template:

“Review the content below. Identify the 3 biggest weaknesses. For each one, explain why it is a problem and suggest a specific fix.

[paste your content]”

This is more useful than asking it to ‘improve’ something, because it forces specific, actionable critique rather than a softened rewrite.

Tips for using ChatGPT for writing

Here are some tips on how to best use ChatGPT for writing to avoid AI-flavored output.

Set the style early on

Before writing tasks, give ChatGPT a style brief. This persists through the conversation so you do not have to repeat tone instructions in every prompt.

Style brief example:

“For this conversation: write in a direct, conversational tone. No corporate language, no filler phrases like “it is important to note” or “in today’s fast-paced world”. Short paragraphs. Active voice. B2B audience: primarily SDRs and sales managers at SaaS companies.”

Use it as an editor

Asking ChatGPT to write from scratch often produces generic content. Asking it to edit or sharpen content you have already written tends to produce much better results – because you are providing the ideas and it is improving the execution.

• “Rewrite this follow-up email to be more direct and cut it by 30%.”
• “This opening line is too soft. Make it more specific to their role.”
• “The ask is buried. Move it to the second sentence.”

Ask for multiple versions

For high-stakes writing – subject lines, CTAs, opening lines – ask for three or four versions rather than one. This gives you options to test and usually surfaces at least one approach you would not have thought of.

Example:

“Write 4 versions of this follow-up subject line. One curiosity-driven, one direct, one that leads with a result, one that asks a question. All under 8 words.”

What not to outsource to ChatGPT

Do not outsource final decisions, unique strategic thinking, or fact-sensitive claims without verifying them. ChatGPT hallucinates – confidently presenting incorrect information as fact.

OpenAI’s prompt engineering guide recommends checking any statistic, quote, or specific claim against a primary source before you publish or send it.

You can ask ChatGPT for the source of its claim and then go there to manually double check.

How to get the most out of ChatGPT for B2B sales and LinkedIn outreach

For sales and lead generation companies, ChatGPT’s highest-value use cases are the ones that compress the time between identifying a prospect and sending something worth reading.

Research and qualify leads faster

Use ChatGPT to synthesise publicly available information about a prospect or company and generate a targeting rationale. Paste in a LinkedIn profile or company description and ask it to assess fit.

Prompt template:

“I am a [your role] at [company]. I sell [product] to [ICP]. Here is the LinkedIn profile of a prospect:

[paste profile text]

Tell me: (1) whether they fit my ICP, (2) what pain points are most likely relevant, (3) one specific personalisation angle I could use in an outreach message.”

This takes under two minutes per prospect and gives you a clear targeting rationale before you write a single word.

This is also where AI lead generation tools that enrich prospect data can pick up where ChatGPT leaves off.

Write outreach messages that don’t sound like templates

The most common mistake with AI-assisted outreach is using output without customisation.

ChatGPT’s first draft is a starting point, not a finished message. The goal is to get 70% of the way there fast, then apply human judgment for the final 30%.

Without direction:

“Hi [Name], I came across your profile and was impressed by your experience. I would love to connect and explore potential synergies between our companies.”

With PTCF prompt and constraints:

“Hi [Name] – saw you recently moved into a growth role at [Company]. We’ve been helping SaaS teams in the [X] space cut their outreach setup time significantly. I did a quick Loom on how – want me to send it over?”

The second version was generated with a prompt specifying role, task, context (job change trigger, SaaS vertical), and constraints (under 280 characters, no pleasantries, reference a specific trigger).

ChatGPT features worth using in 2026

Most users are still on the chat-and-respond loop from 2023. The platform has moved significantly beyond that. These features are available on Plus and above and are directly relevant to B2B uses.

Deep research

Deep Research combs through dozens of online sources – PDFs, articles, research papers – and synthesises a comprehensive report with citations. It takes longer than a standard prompt but produces output that would take hours to compile manually.

Best uses: competitive analysis before a sales call, market research for a new vertical, background research on a target account, pre-meeting prep.

Prompt example:

“Use Deep Research to give me a 500-word analysis of [Company]. Cover their recent product launches, funding history, key leadership changes in the last 12 months, and any publicly stated strategic priorities. Cite your sources.”

Note: Deep research mode takes longer and you should still always double check its claims.

Agent mode

Agent Mode allows ChatGPT to autonomously navigate the web – visiting sites, pulling information, and completing multi-step tasks without you doing it manually. GPT-5.4 is the first mainline model with built-in computer-use capabilities, meaning it can interact directly with software in a build-run-verify-fix loop.

For B2B outreach, the most practical use cases include:

• Researching a target account before a call – company news, recent funding, leadership changes, product launches – without opening 10 tabs.
• Checking whether a prospect’s company is actively hiring in a specific function, which signals growth priorities and budget.
• Pulling recent posts or activity from a prospect’s LinkedIn profile to find a personalisation angle before you reach out.

This pairs well with sales engagement tools that act on the same signals. Give it a specific task with a clear deliverable.

Open-ended requests like “research this company” will produce generic output – “find the last three pieces of content [Name] posted on LinkedIn and summarise the themes” will not.

Projects and custom GPTs

Projects give ChatGPT persistent memory and custom instructions within a defined workspace. 

For sales teams and agencies, this means you can configure a shared environment where ChatGPT always knows your brand voice, your ICP, your product positioning, and your messaging rules – without re-explaining it every session.

Custom GPTs take this further, letting you build a version trained on your specific documents, style guide, and product knowledge. Available on Plus and above; workspace sharing requires Business or Enterprise.

Tasks

Tasks let you set recurring prompts that run on a schedule without you having to trigger them manually. For B2B sales teams, practical use cases include:

• A daily briefing on news in your target vertical – funding rounds, leadership changes, product launches at accounts you are tracking.
• A weekly summary of job postings at your top 10 target accounts, which tells you where they are investing and who the new decision-makers might be.
• A Monday morning digest of LinkedIn content posted by your prospects that week, so you have personalisation angles ready before you start outreach.

To set one up: open a new chat, type your prompt, then click the clock icon in the composer and set the schedule. ChatGPT will run it automatically and surface the results each time.

Available on Plus and above.

Common ChatGPT mistakes to avoid

These are the habits that consistently produce weak output, and the ones worth breaking first.

• Treating it like a search engine. Typing a keyword and expecting a useful answer is not how the tool works. It needs instructions, context, and a defined output format.
• One-shot prompting. Sending one message and accepting whatever comes back misses the point. Iteration is how you get from decent to usable.
• No role, no context. Skipping the persona and background means ChatGPT is guessing at your audience, tone, and purpose. It defaults to generic.
• Taking outputs at face value. ChatGPT can confidently present incorrect information. Any specific claim, statistic, or quote needs to be verified against a primary source before use.
• Sharing sensitive data. Do not paste confidential client information, unreleased product details, or personal prospect data. Review your organisation’s data policy – Free and Go plans allow opt-out of training data use; Business and Enterprise default to no training on your data.
• Using Free or Go for serious work. Neither plan includes GPT-5.4 Thinking, tasks, Codex, or agent mode. For regular business use, Plus at $20/month removes the limitations that make lighter plans frustrating.

How to best use ChatGPT for automated LinkedIn outreach

Before you build a single campaign, ChatGPT can do the groundwork that most people skip.

Use it to:

• Define and pressure-test your ICP – job titles, company size, industry, seniority, common pain points.
• Map out where your audience hangs out on LinkedIn – groups, hashtags, influencers they follow, content they engage with.
• Build your LinkedIn search strategy – which filters to combine, boolean strings to try, Sales Navigator filter logic.
• Identify the buying triggers and pain points to reference in your outreach copy.
• Draft and iterate connection requests, follow-up sequences, and InMail templates using the PTCF framework.

Most teams skip steps 1-3 and go straight to writing messages. That’s why the copy feels generic: it’s targeting everyone and no one at the same time. Once you have your copy and prospect list ready, Expandi handles the execution. It runs sequences, respects LinkedIn’s daily limits, and tracks replies automatically, all in one place.

Expandi’s AI Analyzer categorises those replies into Interested, Maybe Interested, and Not Interested, so you can feed each bucket back into ChatGPT and write targeted follow-ups instead of starting from scratch every time.

Using ChatGPT effectively: FAQ

Making your ChatGPT outreach work at scale with Expandi

ChatGPT gets you to better copy faster. But copy alone doesn’t build pipeline: you need to get it in front of the right people, at the right pace, without getting your account flagged.

Many sales teams using AI for outreach hit the same wall: the messages are sharper, but the execution is still manual. You’re copying prompts into spreadsheets, pasting copy into LinkedIn one message at a time, and losing track of who’s been followed up and who hasn’t.

That’s where the workflow breaks down. 

ChatGPT handles the thinking and the writing. Expandi handles the sending, the sequencing, and the signal tracking. So nothing falls through the cracks and your account stays safe.Start your free 14-day Expandi trial and see what your ChatGPT-built outreach looks like on autopilot.