Low-Quality AI Content: What It Looks Like, How Google Treats It, and How to Fix It

Low-quality AI content is text produced with a language model and published without the work that makes it useful: no original information, no fact-checking, no clear point of view, and no reason for a reader to prefer it over the ten other pages saying the same thing. The problem isn’t that a model wrote the first draft. It’s that nobody did the rest.

That distinction matters, because it’s also roughly how Google frames it. This guide covers what low quality looks like in practice, what Google’s own documentation says (quoted from the source), how to fix a weak draft, and a checklist you can run before anything goes live.

What makes AI content low quality

Most low-quality AI content fails in the same handful of ways. Surface tells (stock phrases, rhythm, over-formatting) are the easiest to spot, and we cover them in detail in signs of AI writing. The deeper problems are about substance.

Nothing new

A model trained on the web reproduces the consensus of the web. Ask it for “tips for onboarding new employees” and it will give you the same tips as every other page, in a slightly different order. If the page has no original example, no data you collected, no process you’ve actually run, it adds nothing to what already ranks.

Unverified facts and invented specifics

Language models generate likely-sounding text. That includes statistics with no source, studies that don’t exist, quotes attributed to the wrong person, and product features a vendor never shipped. A single invented number damages the credibility of everything around it.

Vague, general advice

“Focus on quality”, “know your audience”, “be consistent”. Advice that applies to everything helps with nothing. Low-quality drafts rarely say how, when, with what settings, or when not to.

Mismatched intent

A searcher looking for a template gets an essay. Someone comparing two tools gets a definition of each. The page covers the keyword but not the question behind it.

Padding

Introductions that restate the title, sections that repeat the previous one, conclusions that summarise what the reader just read. Length without information.

No author, no accountability

No name, no relevant experience, no sign that anyone stands behind the claims. For topics where accuracy matters (health, money, law, safety), this is a serious gap.

Volume over value

Hundreds of near-identical pages targeting keyword variations (“best CRM for dentists”, “best CRM for plumbers”, and so on) with the noun swapped and nothing else changed. This is the pattern most likely to run into Google’s spam policies.

How Google treats AI-generated content

Google has published direct guidance on this. Three documents are worth reading in full rather than relying on summaries.

Quality, not production method

Google’s February 2023 Search Central post, Google Search’s guidance about AI-generated content, is headed “Rewarding high-quality content, however it is produced”. It states: “Appropriate use of AI or automation is not against our guidelines. This means that it is not used to generate content primarily to manipulate search rankings, which is against our spam policies.”

The same post draws the line clearly: “Using automation—including AI—to generate content with the primary purpose of manipulating ranking in search results is a violation of our spam policies.”

So AI-assisted content isn’t penalised for being AI-assisted. Content made mainly to rank, rather than to help people, is the problem, whoever or whatever wrote it.

Scaled content abuse

Google’s spam policies define scaled content abuse as “when many pages are generated for the primary purpose of manipulating search rankings and not helping users”. The examples it lists include:

• “Using generative AI tools or other similar tools to generate many pages without adding value for users”
• “Stitching or combining content from different web pages without adding value”
• “Creating many pages where the content makes little or no sense to a reader but contains search keywords”

The phrase that repeats is “without adding value”. A large programme of AI-assisted pages isn’t automatically a violation; a large programme of pages that add nothing is the risk.

Guidance for sites using generative AI

Google’s page on using generative AI content asks site owners to “focus on accuracy, quality, and relevance” and says: “It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing.” It adds that this review “also applies to metadata”, including title elements, meta descriptions, structured data and image alt text.

The helpful content questions

Google’s creating helpful, reliable, people-first content page offers self-assessment questions that work well as an editing brief. Among them:

• “Does the content provide original information, reporting, research, or analysis?”
• “Is this content written or reviewed by an expert or enthusiast who demonstrably knows the topic well?”
• “Does the content have any easily-verified factual errors?”
• “Are you using extensive automation to produce content on many topics?”
• “Are you mainly summarizing what others have to say without adding much value?”

If a draft fails these questions, it is low quality by Google’s own description, regardless of who wrote it.

What this doesn’t mean

AI detectors are not part of this guidance. Google’s documents talk about quality, value and intent, not about detecting model output. Our article on AI content detection covers why detector scores are unreliable as a quality measure.

How to fix low-quality AI content

Fixing a weak draft is mostly adding what the model couldn’t know and removing what it made up. A workable order:

1. Decide whether the page should exist

Before editing, check whether the page serves a real search intent that isn’t already covered on your site. If you have three pages competing for the same query, merging them often beats polishing each. If there is no audience for the page, delete it or noindex it rather than improving it.

2. Re-read the search intent

Search the target query and look at what ranks. Is the searcher looking for a definition, a how-to, a template, a comparison, a tool? Restructure the draft to answer that first.

3. Fact-check every claim

Go line by line. Every statistic, date, quote, price, feature and named study needs a source you’ve opened yourself that says exactly that. If you can’t find one, delete the claim. Don’t soften it to “around” or “many experts say”.

4. Add what only you know

This is where the value comes from:

• Real examples from your work, labelled honestly.
• Your process, with the actual steps, settings and tools.
• Screenshots, templates or worked examples you made.
• Trade-offs and situations where the common advice doesn’t apply.
• Data you collected, with the method explained.

If you have none of these for a topic, consider whether you’re the right site to publish it, or interview someone who has them.

5. Cut the padding

Delete the introduction that restates the title. Merge sections that overlap. Remove the summary at the end if the reader just read the content. Shorter and specific beats longer and general.

6. Rewrite in a real voice

Replace generic phrasing with your brand’s vocabulary and your audience’s words. Vary sentence length and structure. Use lists and tables where the content is genuinely list- or table-shaped, and paragraphs elsewhere. Our AI content editing workflow breaks this pass into stages.

7. Add an accountable author

Put a real name on the page with relevant experience, and have a subject expert review anything where errors could harm the reader.

8. Check the metadata

Following Google’s guidance, review the title, meta description, image alt text and any structured data with the same care as the body.

A worked example (illustrative)

The following is written for this article to show the kind of change involved.

Low-quality draft: “Email marketing is a powerful tool for businesses of all sizes. Studies show that email has an incredible ROI. To succeed, it’s important to segment your list, write compelling subject lines and send at the right time.”

Edited version: “If your list mixes customers and people who only downloaded a guide, send them different emails. A customer needs onboarding and product updates; a guide download needs a reason to try the product. Splitting the welcome sequence by signup source is a one-afternoon job in most email tools and the first segmentation worth doing.”

The draft has an unsourced claim (“studies show”), no specifics and no reason to read it. The edit drops the unsupported claim and gives one concrete, actionable recommendation.

Pre-publish checklist

Run this on every AI-assisted page before it goes live:

• The page answers the search intent behind the target query.
• No other page on the site targets the same query.
• Every number, quote, date and named source is verified and linked.
• Nothing is invented: no fictional studies, testimonials, results or features.
• Illustrative examples are labelled as illustrative.
• The page includes at least one thing a reader can’t get from the top results: an original example, process, template or data.
• Generic advice has been replaced with specific steps, settings or trade-offs.
• Padding is cut: no restated title, no repeated sections, no empty summary.
• Stock AI phrasing is gone (see signs of AI writing).
• A named author with relevant experience is on the page; sensitive topics have expert review.
• Title, meta description, alt text and structured data have been checked.
• Internal links point to genuinely related pages.
• Someone other than the editor has read it once.

Scaling without lowering quality

The risk grows with volume. Ten AI-assisted pages edited by a careful writer are easy to keep useful; a thousand pages generated from one prompt template are not. If you publish at volume, build the checklist into the workflow (briefs with required original inputs, fact-check steps, editor sign-off) rather than relying on individual effort. Our guide to producing AI content at scale covers the process side, and AI content quality control covers review systems.

One of Google’s own warning-sign questions makes a good final test: “Is the content primarily made to attract visits from search engines?” If the honest answer is no, and the page helps the person who lands on it, how the first draft was written matters much less.