Using AI to Generate SEO Content for 250k Product Pages

Will It Hurt Rankings?

[guarriman3]

Hi everyone,

I manage a website that lists 250,000 products, each with its own unique URL (e.g., example.com/p/unique-product-label). The product data comes from a database with fields like product name, manufacturer's name, product dimensions, 5 repair addresses in the USA, 5 URLs to images, price data (2020 vs. 2010), etc.

Currently, each product page starts with a basic intro like: "This is information about [Product Name], manufactured by [Manufacturer's Name]." As you can guess, this is too generic, and Google isn't showing this in the search snippet. Worse, competitors with less content often rank higher than we do. So, I want to improve our ranking and engagement by enriching the content.

My plan is to use the ChatGPT API to generate unique, SEO-optimized text for each product page. I want to programmatically pull the relevant data from our database and pass it to ChatGPT using a consistent prompt, like this:


------------
Create two paragraphs of SEO text for a website about consumer products, focusing on the [Product Name]. The text should be informative, engaging, and optimized for search engines. Use the following data from the database:

Product Name: [Product Name]
Manufacturer: [Manufacturer]
Product Images: [4 URLs of product images]
Repair Locations: [5 Addresses of repair businesses in the USA]
Dimensions: [Product dimensions]
Highlight the unique features of this product, its benefits, and how it compares to others in the market. Mention why it's popular and what makes it a great choice for consumers. Include a call-to-action like encouraging users to view images, check out repair locations, or learn more about the product.
------------

While the API would allow me to scale this for all 250,000 products, I have concerns about potential penalties from Google for AI-generated content. According to Google's guidelines, content that is generated solely for search rankings may violate their policies if it lacks value or uniqueness.

I’m aiming to ensure the content provides real value to users and is well-written, but I won’t be able to manually review each page. Given the scale, I’ll need to rely on the API-generated text for everything.

My questions for the community are:

Does anyone have experience using large-scale AI-generated content for SEO purposes? Were there any noticeable impacts on ranking, either positive or negative?

What best practices can I follow to make sure the AI-generated content is considered valuable by Google, without triggering penalties?

Is there anything I should watch out for or adjust in the plan, especially given that I can't manually review the content for quality?

Any feedback would be really appreciated!

[Whitey]

My questions for the community are:

Does anyone have experience using large-scale AI-generated content for SEO purposes? Were there any noticeable impacts on ranking, either positive or negative?

Yes and no impact - completely neutral for a non-brand. Other factors are more important.

Those multiple observations are taken from using several AI api's, not just ChatGPT, and include extracting data from a unique database set that would be useful to users, giving them information that no competitor has, but , e-commerce brands may behave differently. One significant brand that I'm close to has benefited at scale.

What best practices can I follow to make sure the AI-generated content is considered valuable by Google, without triggering penalties?

In the context of the above I think this is an oxymoron. From a business sustainability point of view anything you can do, Google and brands can do better. Even if you do it better, it doesn't guarantee Google will index all of your pages, or will maintain indexing. AI is a powerful tool, but it may need an innovative and useful angle to prise open a gap that you can exploit imo.

Is there anything I should watch out for or adjust in the plan, especially given that I can't manually review the content for quality?

Maintaining complete quality at scale is not possible. Things will slip through. Apply the 80/20 rule to help break it down, if you can, at each template level. Spot issues here and apply it at scale.

Any feedback would be really appreciated!

My advice, fwiw, would be to go ahead and make your own observations as each experience can be different. I'm not convinced AI will move the needle.

I see evidence that low quality content on branded sites is all that's required. When i say "low quality" i include all forms including poor UI/UX. If 50%+ of your keywords at scale contain a brand name, then all else is largely ignored. AI content writing, in itself won't solve that.

But any good news story is welcome to see, especially one that can positively contradict what i say.

[guarriman3]

Thank you @Whitey for your nice answer.

Your last sentence "AI content writing, in itself won't solve that." may summarize all the test.

The case is that I feel that my webpages may be underranked because they show a few number of words. I prompt just:

- This is X, manufactured by Y
- Dimensions: Z
- links to photos
- etc

But in a very brief and terse way. I do not know if filling the webpage with long adjectives and pronoums would help.

Perhaps the effort of getting more relevant backlinks and obtaining further useful info will help more. Honestly, with 50% of the products (with very low popularity but a very long tail) my website offers the best info of the Internet.

[NickMNS]

My plan is to use the ChatGPT API to generate unique, SEO-optimized text for each product page

If you are going to optimize for anything you should optimizing for users, not for SEO
What does the term SEO-optimized even mean? What will the AI return as a response to that? With 250k products my guess is that the AI will generate text that will be "unique" on a word by word basis but every result will end up being simple variations on the same text. Think back to early 2000, when one would include every possible variation of search term, "large blue widgets", "blue widgets that are large", "large widgets that are blue", etc. Back then they would all be included on a single page, now it will be spread across many. But users will see it for what it is, and googlebot too. If your products are similar then it is normal and expected that the descriptions are as well.

You definitely can use AI to help write generic description, have it add a catchy call to action, but I would not use it for one-off description for 250k products.

If you want to spend loads of money on AI, then create embeddings for all your descriptions, and product images, and add them to a vector database, and let users search for products with the help of AI. A user looking for a "large blue widget" may discover that what they really need is a "round yellow gadget". Better search typically results in better sales. You can then also depend less on Google directing the users to specific pages.

[engine]

I've experimented on a much smaller scale (around 1200 pages) and did find that it became quite repetitive on short text, but then, so does a human.

In fact, I returned to human text because it does need to be unique for the users.

I did not notice any change to ranking on this small scale.