Discussion
  • Content SEO

Where have you seen AI-assisted content at scale work in search, and what was different about the cases that failed?

Published 1 reply

Discussion starter: a content SEO question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.

Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.

1 reply

  • admin
    Admin
    Original poster

    The AI-assisted content that has worked in search at scale shares one feature: each page contains something specific and true that a searcher needs and cannot easily get elsewhere, usually because it is built on the publisher’s own data or expertise, with people checking the result. Product descriptions generated from detailed specifications, local pages built on real inventory or service data, translations reviewed by native speakers, and updates to existing content are typical successes. The failures share the opposite feature: articles generated from keywords, restating what already ranks, published faster than anyone could check. Google has said using AI is not itself against its policies; producing content at scale mainly to manipulate rankings, with little value, is. The difference between the successes and failures maps closely onto that line.

    What separated success from failure

    Cases that worked Cases that failed
    Input Proprietary data, specifications, inventory, expert notes A keyword and whatever the model knows
    Purpose Each page answers a real query with specific information Pages exist to target keywords
    Review Human editing or sampling with quality thresholds Little or none
    Scale Matched to genuine demand As many pages as possible
    Topic fit Within the site’s real expertise Anything with search volume
    Publishing pace Gradual, monitored Thousands of pages at once

    Patterns that tend to work

    • Product and listing descriptions generated from structured data, checked automatically against the source, with brand voice rules.
    • Programmatic pages where each page is built on entity-specific data, such as prices, availability, local statistics or reviews, and AI helps write the connecting text.
    • Translation and localisation with native-speaker review for important pages.
    • Refreshing existing content: identifying outdated facts and drafting updates for editors.
    • Drafting from expert interviews, where the substance comes from a person and AI helps structure it.

    Patterns that tend to fail

    • Blog posts generated for every keyword in a research export.
    • “Ultimate guides” that summarise the top results for a query.
    • Location or comparison pages that swap names in otherwise identical text.
    • Content on topics where the site has no expertise, published to capture traffic.
    • Bulk publishing without monitoring indexing or performance.

    Hypothetical example:

    Two retailers use AI in the same year. The first generates descriptions for 50,000 products from its specification database, with automated checks that every figure matches the source and editors reviewing a weekly sample. Product pages gain long-tail impressions steadily. The second publishes 3,000 AI-written blog posts targeting informational keywords in six months, with no editing. Most are never indexed; those that are rank briefly and then fall, and the site’s established guides lose visibility in the next core update. The model was similar in both cases; the inputs, purpose and review were not.

    Signals to watch while scaling

    1. The share of new pages indexed within a few weeks, by template.
    2. Impressions and clicks per new page, compared with older pages of the same type.
    3. Changes in performance of the rest of the site after large publishing batches.
    4. Error rates found in human sampling.
    5. Engagement and conversion on the new pages.

    If indexing rates are low, new pages get few impressions, or existing content weakens, slow down and investigate before publishing more.

    Watch for this:

    Early success can hide later problems. Some scaled AI content ranks for a while and then falls in a later update, sometimes taking stronger pages with it. Judge programmes over several update cycles, not the first few weeks.

    Governance that keeps programmes on the right side

    Successful programmes usually have a small set of rules written down before generation starts: what data each page must contain, which claims are allowed, what the review sample size is, who can approve publishing, and what metrics would pause the programme. They also keep the programme reversible, with generated content versioned and the ability to unpublish or noindex a batch quickly. That structure matters because the temptation in scaled content is always to publish more, faster. A pause rule agreed in advance, such as stopping if the indexed share of new pages falls below a threshold, is much easier to follow than one invented after traffic has already dropped.

    The question to ask before scaling

    For any AI-assisted content programme, ask: what does each page know that the pages already ranking do not? If the answer is specific data, expertise or a better answer to a real need, AI can help produce it at scale. If the answer is “nothing, but it targets a keyword”, scale only multiplies the problem.

    Use AI to express information you actually have, at the scale of real demand, with human checks and gradual rollout, and avoid using it to produce pages whose only purpose is to target keywords. That is the dividing line between the programmes that last and the ones that fail.

    Need help with this on your own site?

    Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.

    See SEO services

Add a reply

Replies are open to approved members. Log in or apply to join.