codex SEO writing: a complete disaster

bigcoffee

Switched from Claude to Codex for SEO content. spent a month trying to fix it - tweaking prompts, adjusting system instructions - but the output is still junk. Robotic, unnatural, zero readability. The core problem? Codex isn't built for writing. Its system prompt is entirely code-focused. No instructions on tone, structure, or word choice. You can try a custom subagent with a dedicated copywriting prompt, which helps marginally, but the underlying model just doesn't do long-form SEO well. AI writing tools are only as good as the prompts, sure, but some models are fundamentally wired for something else. Codex is one of them

roichaser

Honestly, this feels like a cop-out. Blaming the system prompt for Codex's garbage SEO writing? That's like blaming the keyboard for a bad article. The tool is built for code, full stop. If you're shoving it into content workflows without a custom prompt, you're the problem, not the model.

I've run my own subagent with a tailored copywriting system prompt - still a struggle. The real issue isn't the prompt, it's that AI fundamentally lacks the nuance for SEO writing. It can't discern "valuable" from "fluff." It spits out keyword-stuffed robot vomit because that's what its training data rewards. You can tweak prompts till the cows come home, but the core output's still a glorified Markov chain.

So yes, Codex produces atrocious SEO writing. But so does GPT-4, Gemini, or any other model if you're naive enough to trust it with your content strategy. The game's rigged from day one.

trendspotter

watching AI churn out that painfully generic sludge makes me cringe. it's like cheap polyester pretending to be silk, no texture, no soul.

dusting

AI writing mirrors the prompter's clarity. but some models default to a sterile tone, like a corporate memo ghostwritten by a spreadsheet

budgetburner

I think for any AI, lack of proper prompt in explaining the right structure, can be detrimental and impact negatively for seo

brandvoice

Oh, this resonates. I've watched so many clients throw money at different AI writers, convinced the next one will save them. But honestly? It's nearly always the same problem underneath. The tool itself isn't the enemy - it's how we're asking it to write.

If you feed it vague instructions, it'll churn out safe, soulless filler every time. Switch to another platform and you just swap one flavour of generic for another - verbose rather than stiff, but still completely forgettable. What actually matters is giving it a tight structure, a clear intent, and real examples of what "human" reads like for your audience.

When the output feels consistently junk, nine times out of ten it's a prompt design issue or a lack of post-editing rules. The model can only work with what it's given.

harper-local

anyone found decent alternatives? asking for a mate

logan-lifecycle

Honestly, I've seen this play out across dozens of customer journey audits. codex and its ilk crank out what looks like keyword-dense filler, but they completely miss the voice triggers that keep users in a nurture flow. If your win-back sequences rely on that stuff, you're just adding noise to an already bloated segment.

Producing your own writing isn't just about quality-it's about mapping the right tone to each journey node. automated copy can handle the low-involvement branches (cart reminders, basic re-engagement), but for the high-intent segments you need actual human signal. otherwise you're optimising the wrong logic branch

pixelperfect

Gemini and Sonnet 4.6 have worked well for me.

zoecreative

Honestly, Codex is built for code, not copy. For SEO writing, I'd take raw content, drop it into ChatGPT or Claude, and prompt for structure and keyword optimisation. still needs a human to keep it from reading like a robot's shopping list.

maya-b2b

I have Claude writing prompt for codex to write content. I likely depends but gpt 5.5 subscription seemed the best value for SEO content for me. I don’t want to waste my 200$ Claude subscription on SEO content as I hit subscription limits some times but no question I think Claude is better at top quality everything. I use a lot of all models including gpt 4 mini. Right model, right task, right prompts.

bigcoffee

Honestly, I'm with you. I've wasted too many hours tweaking prompts just to get Codex to stop vomiting keyword-stuffed drivel. Conversion rates don't lie - I tested its output against Claude for a B2B nurture sequence and saw a 40 % drop in click-through from the Codex version. Back to Claude it is.

budgetburner

Codex isn't the real villain here. The problem is that every AI model - Perplexity, ChatGPT, Gemini - is scraping the same mediocre Reddit threads because Google keeps ranking them. Garbage in, garbage out.