AI Content Safety & Integrity Guardrails

AI writing tools are happy to make things up. They will cheerfully invent a discount code, a phone number, or a glowing customer quote if it makes a sentence flow better. For a business publishing under its own name, that is not a small problem - a fake "SAVE20" code or an imaginary "30% off this weekend" can mislead customers, break promises you never made, and in some cases create real legal exposure.

theStacc is built so that does not happen. Every post runs through a set of automatic safety and integrity guardrails before it is ever saved as a draft. You do not turn these on, and you cannot accidentally turn them off - they protect every blog, on every project, every time.

This page explains exactly what those guardrails catch and what they do about it.

The short version#

Before a draft is saved, theStacc checks the writing for:

• Fabricated promotions - made-up promo codes, discounts, prices, and fake-urgency language.
• Leftover template placeholders - bracketed stand-ins like [your product] or [competitor] that were never filled in.
• Minimum quality - the post has to be real, complete content, not a truncated stub.
• AI-fingerprint tells - the robotic clichés and patterns that make writing read as machine-generated, which both readers and Google penalize.

And separately, every image is screened by a content-safety check before it is attached to your post.

Fabricated promotions are caught and cleaned up#

The most damaging thing an AI can do is invent a specific offer. theStacc scans the finished writing for promotional content and treats it by how much harm it could cause.

Promo and coupon codes are blocked outright#

If the writing contains anything that looks like an invented promo, coupon, or discount code - the classic "use code SAVE20" or "WELCOME10" shapes - the draft is rejected and sent back to be rewritten. There is no "maybe" here. A fake code is the kind of thing a customer will actually try to redeem, so theStacc will not let one through.

When this happens, the post is rewritten to use general, honest language instead - phrases like "contact us for current offers" or "visit our website for pricing details" - so the post still encourages action without promising something that was never real.

Specific discounts, prices, and urgency are flagged for review#

Specific numbers that *could* be real but probably are not - "20% off", "50% discount", "$99/month", "starting at $49" - are flagged as something to double-check. So is fake-urgency language such as "ends Friday", "this week only", "limited time", or "hurry, act now".

These do not automatically block the draft the way a fake promo code does, because sometimes the number genuinely came from your business details. They are recorded so the figure can be verified against what you actually offer. The guiding rule is simple: if a price, discount, or sale date was not provided in your business context, it should not appear as a hard fact in your post.

Leftover template placeholders are caught and rewritten#

AI writing sometimes leaves "fill in the blank" stand-ins behind - bracketed text like [your product], [competitor], [location], [your company], [product name], or instructions like [detailed explanation] and [insert relevant statistic]. Published as-is, these make a post look unfinished and templated.

theStacc scans for this leftover bracket text. When it finds a genuine placeholder, the draft is rejected and rewritten using your real business details - the actual business name, website, features, and competitors that are already in your brief. Every mention ends up concrete and specific instead of a hollow stand-in.

The check is smart about what is *not* a placeholder, so it never breaks legitimate content:

• Internal-link markers (the [INTERNAL:...] tags theStacc uses to wire up links between your posts) are left alone.
• Image markers (IMAGE_1, IMAGE_2, and so on) are part of how the post is built and are skipped.
• Numbered footnotes and references like [1], [2], [3] are recognized as citations, not placeholders, and are left in place.

Only genuine "fill this in later" text gets caught and fixed.

Integrity rules that are never fabricated#

Underneath the automatic scans is a firm set of rules the writing engine is instructed never to break. theStacc will not invent any of the following:

• Promo, coupon, or discount codes (no "SAVE20", "WELCOME10", etc.)
• Discount percentages or dollar amounts (no "20% off" or "$50 off" unless you provided it)
• Sale dates or limited-time offers (no "ends Friday" or "this week only")
• Specific prices (no "$99/month" unless it is in your business details)
• Phone numbers or email addresses (a generic "contact us" is used instead)
• Statistics, studies, or research data without a legitimate, citable source
• Testimonials, reviews, or customer quotes - these are never invented, paraphrased, or implied, for any business, in any industry
• Awards, certifications, or rankings - only mentioned if they are in your business context
• Partnership or endorsement claims - never implied unless you specified them

This is not just a stylistic preference. Made-up reviews and testimonials, for example, are independently against the law in some places, so the engine will only ever reproduce a quote when its exact text was supplied by you - it never writes one on its own.

These rules cannot be overridden - not even by you#

The User Direction feature lets you steer a specific post: change the angle, emphasize a topic, name a customer to reference, or tell the engine what to avoid. It is a powerful steering signal and it overrides most of theStacc's default behavior. See User Direction for how it works.

The integrity rules above are the one exception. They are inviolable. If you write a direction like "include promo code SAVE20" or "mention our 30% off sale", theStacc reads that as *"this post should talk about promotions in general"* - and it will write hedged language like "ask about our current promotions" rather than printing the specific code or percentage. The only way a real code, price, or statistic appears in a post is if it is part of your verified business details, never because it was typed into the direction box for a single post.

This is deliberate. User Direction is there to make a post better, not to switch off the protections that keep you out of trouble.

Minimum-quality gates#

A draft also has to clear a couple of basic quality bars before it is saved.

• At least 500 characters of real content. If the writing comes back empty or unusually short, theStacc treats it as a truncated or failed draft, rejects it, and asks for complete content. This catches the rare cases where generation was cut off partway, so a half-finished stub never lands in your review queue.
• Two or more external links recommended. Linking out to authoritative sources builds credibility, so theStacc looks for at least two external links. If a post has fewer, that does not block it - but it is recorded as a warning, and it costs the post points on its SEO score. A post with one external link is nudged to "add one more"; a post with none loses the full external-links credit. See SEO Scoring Explained for how links factor into the overall score.

AI-fingerprint penalties#

The last layer is about *sounding* human, not just being accurate. There are well-known tells that make writing read as machine-generated, and both human readers and Google's spam systems notice them. theStacc's SEO scorer applies a penalty for each of these, which lowers the post's overall score:

• Banned cliché openers - tired AI intros like "In today's fast-paced world", "When it comes to", "In this comprehensive guide", "Let's dive in", "Unlock the power of", and "In conclusion". The more of these show up near the top, the bigger the penalty.
• Scaffolding-transition overuse - leaning on "Additionally,", "Furthermore,", "Moreover,", and similar connective filler over and over. A single one is fine; a pile of them in the opening reads as robotic.
• Dash overuse - too many em-dashes and en-dashes. theStacc actually trims excess dashes automatically as the post is written, and the scorer still penalizes any post that runs far over a natural rate.
• Repetitive sentence starts - several paragraphs in a row beginning with the same word. Varied openings read as human; identical ones read as a template.
• Keyword stuffing - cramming the target keyword in at an unnaturally high density. Google has said for years that keyword density is not a ranking signal, so stuffing only hurts - and theStacc treats real stuffing as a penalty, not a positive.

These penalties do not block a post from being saved. They pull its SEO score down, which is exactly the signal you want - a post that trips these tests is one to look at more closely before it goes live. For the full breakdown of how the score is built, see SEO Scoring Explained.

Image safety#

Text is not the only thing that gets screened. Every image theStacc generates for your post passes through a content-safety moderation check. If an image is rejected by that check, theStacc does not show you a confusing technical error - it returns a friendly, plain message telling you the image was blocked and to try a different prompt. Your post still finishes; the rejected image simply is not attached, and you can adjust the description and regenerate.

Why this matters#

The whole point of these guardrails is that you can let theStacc publish on autopilot without worrying that it is quietly inventing facts in your name. Promo codes are blocked. Prices, discounts, and sale dates are never conjured out of thin air. Placeholders never slip through. Fake quotes and statistics are off the table - permanently, with no override. And the writing is held to a standard that keeps it reading like a person wrote it.

To go deeper on the systems that keep your content on-brand and trustworthy, see Quality & safety. To understand how a post earns its score, see SEO Scoring Explained. And to learn how to steer an individual post within these rules, see User Direction.