6 Best Practices to Follow When Using AI to Generate Content

Generative AI often gets a bad rap as a tool to help produce written content. Unfortunately, some of the bad rap is warranted. 

However, AI can provide real benefits to your content strategy. It can help with research, conduct content audits, analyze existing voice and tone, and save time with (really) short-form content. 

I like using ChatGPT to generate social media posts for blog articles and e-books that I’ve written. The posts aren’t perfect out of the gate, but it gives me a good head start to massage them into something not just usable but also highly effective.

Long-form content and digital storytelling, however, don’t come as easily for generative AI. To quote Eric Clapton, “It’s in the way that you use it.” Following these six best practices, AI can be used—really carefully and patiently—to develop, generate, and finalize content:

1. Be Realistic

Unfortunately, many people overestimate the capabilities of large language models (LLMs), thinking they can deliver perfect, impactful copy with simple prompts. Generative AI simply can’t—read that again, it can’t! However, AI can help non-confident writers get content started, organize notes, and overcome mental blocks and copy obstacles. For example, if you write something that you just know is blowing chunks, asking AI to rewrite is a great launchpad to help you work through to something readable.

Whatever strategy you adopt when using AI, don’t expect that you’ll get all the copy you need in a few minutes, with no additional work. To make the most of LLM content, you should be spending as much time as needed to review, edit, and perfect the copy. That leads us to a crucial next step … 

2. Check Everything

In May 2025, a freelance writer used AI to generate a list of 15 books for summer reading and submitted the content to his syndicator. The non-bylined list was picked up and published by the Chicago Sun-Times. There was one massive problem with the list—10 of the titles were fake, made up by the generative AI, which the writer admitted to using (reportedly, just for research) without double-checking the work.

No one at the Sun-Times caught the mistake (calling this a “mistake” is being generous …), either, and a printed version of fake books was delivered to people’s doorsteps. The failure provides a valuable lesson: Double-check everything that generative AI spits out for you. Don’t skip this step, and don’t be lazy—your reputation is at stake. This might cut into the time savings you’re hoping an LLM is supposed to provide, but an extra hour now can save you hours in damage control later.

3. Trust Yourself, not the AI

Along the same lines as fact-checking LLMs, if generative AI is suggesting something that contradicts your knowledge, experience, and/or every fiber of your being, ignore the suggestion! Don’t even think twice, and don’t doubt yourself. 

Moreover, tell the AI why you think its suggestion is a bad idea. Theoretically, it should learn and not repeat the mistake. That doesn’t mean you implicitly trust it the next time, but at least, maybe, subsequent reviews become a little easier.

4. Detailed Prompts

LLM-generated content is only as good as the prompts you direct AI to follow. Give it a one-sentence prompt that’s open to interpretation, and who knows what you’ll end up getting back. Give detailed instructions that include what you want for topic, structure, voice, tone, audience, sources, length, keywords, attitude, and style—and you’re more likely to receive copy that will require less editing. 

Of course, good prompting takes time and practice. Having some basic prompts (e.g., a detailed paragraph on your brand’s voice and tone) on hand that you know work helps, but be prepared to spend several minutes, at least, explaining to the AI exactly what you’re looking for. 

Also, when copy is generated, don’t be afraid to prompt back suggestions for the AI to fix things you don’t like or it did wrong. Although you risk the AI making things worse, if the copy wasn’t meeting your minimum expectations, you don’t have much to lose. 

5. Be Prepared to Rewrite … a Lot

Here’s where the hard—and essential—work comes into play when using generative AI. Take the framework of what AI returns and aggressively revise it into something that appeals to readers, matches your brand, is accurate, and won’t embarrass you or your organization. These editing steps will get you started:

• Revise the introduction so that it grabs the reader’s attention, clearly sets up the narrative, and gets to the point quickly.

• Ensure keywords are used effectively and not forced.

• Edit up or down to match your desired word count.

• Rewrite H2s and H3s in case they, well, totally stink.

• Ensure any facts presented are accurate and come from reputable sources.

• Add external and internal links if the AI didn’t.

• Revise the conclusion so that it’s not regurgitating the content.

• Eliminate cliches and AI-favorite cliches and expressions.

• Take out excessive em dashes (not more than one or two a paragraph) and other constructions that AI over-relies upon.

• Ensure paragraphs are of appropriate length for the content you’re writing.

• Remove repetitious copy and tighten the flow of the narrative.

• Add crucial things that the AI missed, and remove any obvious or non-helpful advice.

Whew! That’s a big list for sure, but you might take the same steps if you were editing a co-worker’s or contractor’s copy. Keep these two things in mind as you edit:

1. One of your editing goals is to make the content sound as minimally AI-generated as possible. Although AI content checkers are hit or miss on their accuracy (I’ve seen them flag human-written copy as AI), the last thing you want is for a reader to run your content through one and identify it as AI. 

2. If you are unsure of the quality of the copy after you’ve completed your rewrite—or just don’t feel confident in your editing skills—run it through Grammarly or back through the AI.

6. Cut the AI Off

At some point, no matter how many prompts and replies you throw at the LLM, and no matter how much you beg it to improve the copy, you may need to accept that generative AI has taken the content as far as it can go. When that happens, cut the AI off and work the copy the rest of the way. 

Here’s the potentially frustrating part: You may need to accept that the LLM failed and start over. Remember, generative AI, for all its promise, is far from perfect—it can’t even defeat a 1978 Atari chess cartridge, so expecting it to deliver stellar marketing content every time is unrealistic. Don’t waste more time and brain cells trying to fit the square AI peg into a round content hole.

Generative AI may be the future of marketing, or its future may be uncertain amid plenty of backlash. Whatever looms, it can’t be ignored. Whether AI is a big part of your content strategy or barely registers, you should have some sort of policy in place defining how you use AI and the best practices that should be followed as part of that usage.

If you don’t want the hassle of using generative AI for content, consider turning to the experts at Unretained. We can fill the content gaps in your marketing and sales enablement strategy with human-driven blog articles, web copy, e-books, and sales collateral. Let’s talk!