Why Your ChatGPT Prompts Aren’t Working (And How to Fix Them), According to HubSpot’s SEO Team

There's a batch of hype astir ChatGPT and AI models for illustration it. It tin return a azygous punctual and toggle shape it into an wholly caller portion of content. It tin investigation virtually immoderate topic, and present its findings successful nan shape of poetry, code, lyrics, and more.

And yet, erstwhile you occurrence up ChatGPT and provender it a prompt, it responds pinch something… acold from perfect. It's excessively broad, excessively long, excessively bland, aliases misses nan people entirely.

This often happens erstwhile we don't cognize really to ask, successful AI-terms, what we want. In different words, nan output is only arsenic bully arsenic nan input. So, if you want to maestro AI, knowing really to constitute a bully punctual is key.

Here, I said to HubSpot's SEO squad to understand nan cardinal elements of a bully prompt, on pinch immoderate adjuvant punctual penning tips and techniques.

Table of Contents

How to Write nan Perfect AI Prompt

Prompt Writing Techniques

How to Write nan Perfect AI Prompt

1. Envision your perfect response.

It's tempting to load up ChatGPT and jump correct into nan prompt. However, Josh Blyskal, Associate SEO Strategist astatine HubSpot, recommends a much introspective approach.

"The first point you should do is fig retired what you want. Before I type anything, I ever return a infinitesimal to envision what nan last output will look like," he told me.

He gives an illustration of a marketer who wants to make a trading plan. Instead of typing "Write maine a trading plan" into ChatGPT, she should return nan clip to envision what nan scheme looks for illustration — really galore sections it has, really galore slug points are nether each section, and what accusation it contains.

Now that she has a amended grasp of nan task astatine hand, it becomes overmuch easier to constitute a punctual that checks each nan boxes.

2. Write a specific, action-oriented task.

AI models thrive connected specificity. If you provender it pinch vague instructions, you springiness it room to construe what it thinks you want.

For example, suppose you punctual ChatGPT to "Rewrite this [article/email/Tweet]," expecting it to constitute thing caller and original. Instead, it regurgitates your original connection utilizing nan aforesaid words and phrases.

The solution is to get circumstantial and intentional pinch your words.

Instead of "Rewrite," for instance, you mightiness opportunity reimagine, expand, simplify, aliases modernize. Even a mini alteration successful nan measurement you building a punctual tin importantly alteration nan response.

As Blyskal points out, "A batch of group dainty AI for illustration a rigid strategy that tin beryllium hacked pinch nan correct words aliases phrases, but that's not nan case. There are nary concealed keywords that tin guarantee nan champion results. It's for illustration playing a crippled of words, and you person to deliberation intentionally astir nan ones you use."

3. Set nan shape pinch context.

You could get distant pinch stopping astatine nan erstwhile measurement — particularly if your petition is reasonably straightforward (e.g., "Write 5 knock-knock jokes for kids").

However, you tin supercharge your prompts by padding them pinch context. This gives nan AI exemplary much ingredients to navigator up a awesome response.

Bianca D'Agostino, SEO Specialist astatine HubSpot, puts it this way: "Suppose your friend asks you to spell to nan shop and prime up a fewer things. You decidedly request much discourse – which store, what things, what's nan budget, and truthful on. The much discourse you get from your friend, nan easier it is to do nan task. The aforesaid goes for AI!"

Let's look astatine an example:

Both of these prompts mightiness output bully results, but nan 2nd punctual will springiness you a much elaborate and compelling description. This is because nan AI exemplary has much discourse to activity with, including nan product's sanction ('Squeaky Queen'), its shape (a scrub), and its cardinal features (vegan, gluten-free, recyclable packaging).

4. Add clear instructions.

Although Al models are powerful, they still request you to return nan lead. Without clear instructions, they're fundamentally a cook without a look aliases a caller worker without a handbook.

Here's an example: Imagine you request to constitute an article astir Instagram marketing. Along pinch nan main task ("Write an article astir Instagram marketing"), you besides instruct ChatGPT to:

• "Cover nan pursuing sub-topics…"
• "Incorporate nan pursuing quote/statistic…"
• "Add an illustration to demonstrate…"

Not only are you giving ChatGPT a clear topic, but you're besides guiding and shaping nan contented itself. You're adding quotes, statistics, and examples to springiness nan output much texture.

Of course, it's not conscionable what you opportunity but really you opportunity it. The contented of nan article mightiness beryllium up to par, but it whitethorn deficiency a unsocial sound aliases perspective. This is wherever style and tones travel in.

Adding Style and Tone

For Sylviain Charbit, HubSpot Senior Technical SEO Specialist, adding style to prompts is simply a multi-step process. He starts by letting nan AI instrumentality make a wide reply and past progressively molds its consequence for style.

"You tin inquire it to animate itself from a situation, character, aliases profession, aliases supply it pinch already made examples to emulate," he told me. Charbit besides incorporates reside modifiers, specified arsenic "cheerful," "humorous," aliases "authoritative," to guideline nan AI further.

Let's look astatine immoderate examples:

Other style parameters see magnitude and format. You tin specify nan magnitude of nan consequence (e.g., "Keep nan article nether 800 characters") and its format (e.g., "Create a database of slug points").

5. Clarify and Refine

Blyskal follows nan 60% Rule.

He told me, "When you make thing and it's astatine slightest 60% of what you want, you should proceed refining that circumstantial prompt. But if it falls short of that, it mightiness beryllium clip to commencement from scratch."

So, if you inquire ChatGPT to make 10 blog ideas and you only for illustration two, it's a bully denotation to spell backmost to nan drafting board. Even if you meet nan 60% threshold, you tin still refine nan output by giving follow-up instructions specified as, "Make it sound much conversational" aliases "Condense this into 1 paragraph."

Remember that punctual penning isn't an nonstop science. Like immoderate imaginative process, it often takes aggregate revisions to get nan correct output.

Prompt Writing Techniques

Give It a Role

Role-playing is simply a method that involves asking nan AI exemplary to adopt a circumstantial role, job, aliases function.

"When you inquire an AI exemplary to play a circumstantial role, you're giving it a definite level of expertise," Blyskal told me. "For instance, if you punctual it to enactment arsenic a master interviewer, it will look for examples of bully interviewing skills and respond pinch professional-seeming questions."

By defining a domiciled for nan AI exemplary to play— specified arsenic a master interviewer, a schematic designer, aliases a nutritionist — you're besides tapping into different viewpoints and perspectives, which tin beryllium particularly adjuvant for marketers.

For example, let's opportunity you're launching a caller product, a acold plunge tub. You request to create an advertisement that appeals to your target audience: specifically, long-distance runners. In this scenario, you inquire nan AI exemplary to play nan domiciled of your target assemblage and database nan cardinal benefits of your product.

Now nan AI exemplary will make responses that align pinch nan needs and preferences of long-distance runners. This gives you valuable penetration into your audience, enabling you to create a much compelling and applicable ad.

Chain of Thought

Imagine you're connected a roadworthy trip, and you're relying connected GPS to guideline you from constituent A to constituent B. Throughout nan journey, it gives you turn-by-turn directions until you scope your destination.

Chain of thought prompting follows a akin approach. You springiness nan AI exemplary 1 punctual aft another, pinch each punctual building connected nan 1 earlier it. This keeps nan exemplary connected track, giving you greater power of nan output.

"You're fundamentally asking nan AI exemplary to explicate its reasoning astatine each step. If you don't for illustration immoderate of nan explanations, you tin tinker pinch nan output until it's fresh for nan adjacent prompt," Blyskal explains.

This is perfect for tasks that tin beryllium surgery down into smaller steps. For example, see a copywriter utilizing AI to create an article. She first prompts it to create an outline. Once she's happy pinch nan outline, she prompts it to constitute nan article, and past nan headline, and past nan meta description. She doesn't move to nan adjacent measurement until she's satisfied pinch nan existent output.

That's nan powerfulness of concatenation prompting. By guiding nan exemplary successful a logical sequence, you make much meticulous and applicable responses.

Provide Examples

"If there's immoderate 1 extremity that will return your prompting to nan adjacent level, it's providing examples to nan AI model," Blyskal told me.

You tin deliberation of examples arsenic reference points, helping nan AI exemplary emulate a definite style, tone, structure, aliases format. On apical of that, AI tin analyse a portion of contented and connection improvements.

Back to You

Blyskal said it best: "At nan halfway of prompting is conversation." You don't technologist aliases programme AI. Instead, you talk to it.

Like immoderate conversation, you people set your tone, supply ample context, and stock details, anecdotes, and examples to springiness your communicative texture. These are nan aforesaid elements that make a really awesome prompt.