One AI writing style ruins both recipes and product pages because the two sit at opposite ends of why someone is reading — a recipe teaches you to cook something, a product page helps you decide to buy something, and a voice tuned for either reads wrong on the other. Feed your AI only recipe posts and the product pages turn flowery and bury the price; feed it only product copy and the recipes stiffen into ad slogans. The fix isn't a smarter single prompt — it's a separate learned style per post type.
We kept hitting this on the most ordinary kind of WordPress site: a food blog running WP Recipe Maker that also sells a cookbook or a few spice tins through WooCommerce. Before we split voices by post type in Contexta, a model handed the whole site as one blob wrote to the average of it — and on a site that's ninety-percent recipes, the dozen product pages came out faintly scented with cinnamon and nostalgia. That averaging is the whole problem, and it runs in both directions.
Why does one AI writing style ruin both recipes and product pages?
Because a single style can only match one search intent, and recipes and product pages don't share one. A recipe answers "how do I make this"; a product page answers "should I buy this" — different reader, different mood, different next click. A voice that nails one is, by construction, wrong for the other, so "use our brand voice everywhere" quietly means "use the wrong voice on half the site."
The mechanism underneath is averaging. An AI with no per-type signal writes to the mean of the examples it's shown, which is why generic output reads so uniform — it's the average of your whole site's voice rather than the register a specific page needs. Give it two hundred recipes and a dozen products and the mean is a recipe, so the products inherit that cadence. Flip the ratio and the recipes harden into spec-sheet prose.
Neither result is "bad writing" in isolation. It's competent prose pointed at the wrong job — which is harder to catch, because nothing looks broken until you notice the product page never actually says what the thing costs.
What does each page type actually need from its writing?
A recipe needs a short, warm lead and then the method front and center — numbered steps, exact quantities and timings, marked up as a recipe so Google can build the rich result. A product page needs almost the reverse: the answer first (what it is, what it costs, why it's the right one), trust signals, and a single clear call to action. Same site, two nearly incompatible briefs.
Both still obey the same sentence-level rules we cover in GEO vs SEO writing — answer first, name your entities, make each passage stand alone. What differs is what the answer is and what tone carries it. A cook wants grams and minutes; a shopper wants price, fit and proof. The structure rhymes across the two; the voice can't.
The axes that pull them apart:
- Intent — learn and cook, versus decide and buy.
- Structure — a story then the numbered method, versus the answer then specs and a call to action.
- What the AI lifts — your ingredient list and steps for a recipe rich result, versus price, availability and specs for a shopping answer.
- Success — saves, prints and dwell time, versus add-to-cart and checkout.
Miss any one of these and the page still publishes; it just quietly underperforms at the only thing it was there to do.
What breaks when you apply the wrong voice to each?
The wrong voice doesn't just read oddly — it removes the thing each page exists to deliver. A recipe written in product-page voice swaps the method for superlatives; a product page written in recipe voice smothers the price and specs under a nostalgic lead no buyer reads. Map it out and every single style covers one diagonal and wrecks the opposite corner.
↑ Voice the AI learned
Recipe voice → recipe post
Works: warm one-line lead, then the numbered method and recipe schema up front.
Recipe voice → product page
Breaks: flowery intro buries the price, the specs and the add-to-cart button.
Product voice → recipe post
Breaks: "premium, artisanal, best-in-class" where a cook just wants grams and minutes.
Product voice → product page
Works: answer-first spec, trust signals, one clear call to action.
Page it's used on →
The product-page corner is the more expensive one to get wrong, because a recipe-trained model loves to embroider — and an embroidered product page is exactly where AI writing starts inventing details it shouldn't, from materials and dimensions to prices that don't match your catalog. A recipe that's slightly too salesy loses a little dwell time; a product page that's slightly too creative can state a wrong price to a shopper about to buy.
Isn't a single "brand voice" prompt supposed to handle this?
A brand-voice prompt sets personality — playful, formal, warm — but personality isn't register, and register is what changes between a recipe and a product page. You can be the same brand in both places and still need cooking prose here and buying prose there; one prompt can't hold both without collapsing back to the average. Brand voice answers "who's talking"; it doesn't answer "what is this page for."
It's the same reason one voice can't cover two languages: each language needs its own learned profile, because a translated voice inherits the wrong rhythm. Post type is just another axis the profile has to split on. Treat "recipe" and "product" as two dialects of your brand, not one voice stretched across both.
The honest caveat: if your site is all one type — a pure recipe blog, or a pure shop — a single voice is genuinely fine, and splitting profiles is over-engineering you don't need. The split earns its keep the moment you truly span both, which on WordPress usually happens the day a content site bolts a WooCommerce shop onto the side.
How do you keep a separate style per post type without maintaining a dozen prompts by hand?
You let the tool learn a profile from each post type's own published content instead of writing and versioning prompts yourself. The manual version — one carefully tuned prompt per type, kept in sync as your voice drifts — is exactly the kind of upkeep that survives about three months before everyone quietly falls back to one generic prompt.
Contexta's AI editor learns a separate voice per post type (and per language) from your existing posts, so a recipe is generated in your recipe register and a product page in your product register — one engine, two learned styles — while it pulls real prices and specs from your catalog so the product side stays factual rather than embroidered. You're not maintaining prompts; you're pointing it at content you already wrote.
Find the split first: list your post types and notice which ones sit on opposite sides of buy-versus-learn. Then give each the voice its readers actually came for. One style was never going to serve a cook and a shopper at the same time.
FAQ
Can I use the same AI tool for recipes and product pages?
Yes, one tool is fine as long as it writes each in a separate learned style per post type — the problem isn't the tool, it's using one voice for both. A single global prompt averages the two registers together, so a recipe ends up salesy and a product page ends up flowery. On a pure single-type site one voice is genuinely enough; the split only matters once you span both recipes and products.
Why do my AI product descriptions sound like blog posts?
Because the model learned from a site where blog or recipe content outnumbers products, so it writes product copy to the average of your whole site, which leans editorial. The fix is a product-specific voice learned only from product content, plus real catalog data so it states specs instead of embroidering them. Check whether your tool trains one global voice or a separate one per post type.
Does a recipe blog with a shop need two writing styles?
Yes — the day you add a WooCommerce shop to a recipe blog you've created two opposite content types on one domain, and one voice will read wrong on at least one of them. Recipes need method-first cooking prose with recipe schema; product pages need answer-first buying prose with price and specs. Keep the structural discipline shared and let the register differ between the two.
Will writing product pages in my recipe voice hurt sales or SEO?
It can hurt both: a flowery, story-led product page buries the price, specs and call to action that convert, and it gives an AI writer room to invent product details that don't match your catalog. Buyers and AI shopping agents both want the facts first, and a recipe voice delays them behind a nostalgic lead. Write the product page answer-first and save the storytelling for the recipe.