The 'SEO Heist' Revisited: What Followed an AI-Copied Competitor Sitemap

• The method, as one source tells it, was to export a competitor's sitemap, turn the URLs into titles and generate about 1,800 articles with AI.
• The reported results are 3.6 million views in 18 months and 489,509 sessions in a single October, with a later 42% fall in organic traffic.
• No figure is dated precisely, no metric is defined, and the cause of the fall is not shown, so none of it is an audited result.
• Google's current guide treats separate pages for query variations, made to manipulate rankings, as a spam policy violation.

What was the SEO heist?

Spring Agency's account names an SEO company founder, Jake Ward, as the person behind it. It says he diverted 3.6 million views in 18 months from a competitor to his own target site.

This page does not repeat the story as a legend. It separates what the supplied sources state from what a reader would have to infer, and it marks where the public record, as supplied here, stops.

How did the method work?

According to Spring Agency, the method had three steps and rested on an AI tool and a competitor's public sitemap. Each step below is Spring's description, not something observed directly.

1. Export the competitor's sitemap, the public list of its page addresses.
2. Convert each address into an article title.
3. Generate about 1,800 articles from those titles with AI.

Spring says this drew Google's attention away from the original site and toward the new one. That is a causal claim, and the evidence supplied does not show the mechanism.

Several details a reader would need are missing. The AI tool, the prompts, any human review, the publishing pace, the domain names and the competitor's identity are not in the supplied text, so this page does not name them.

What figures were reported, and how firm are they?

Five figures carry the story, and none of them is a measurement a reader can check. The table lists each one with its source and the reason it cannot be treated as audited.

The wording of the headline number varies across the search results listed in the evidence. Spring writes “views”, the title of a result on X says “total traffic” and the title of a Medium result says “visits”. Views, visits, sessions and traffic are different units, so figures in one unit cannot be compared with figures in another.

What does the arithmetic show?

The arithmetic shows that the reported numbers are at least consistent with a steep climb, and that the 42% fall cannot be read without a baseline. The table derives four figures from Spring's numbers, with the inputs shown.

3,600,000 ÷ 18 = 200,000 a month; 489,509 ÷ 2 ≈ 244,755; 489,509 × 0.58 ≈ 283,915

The last row is the instructive one. If the 42% fall were measured from the October peak, traffic would still sit above the level six months before it. If it were measured from some other month, the result changes, which is why the baseline matters more than the percentage.

These sums also mix units, because the 3.6 million is described as views and the 489,509 as sessions. Treat the table as a way to see what is missing, not as a finding.

What is the timeline, and which dates are sourced?

Only the background dates are sourced, and the dates of the case itself are not in the evidence. The table below was built by taking every dated or sequenced event in the supplied text and labelling how firm each date is.

The Business Insider address in the search results carries a December 2023 date, which makes 2023 the likely year of the case. A reader who needs the exact days must go to the primary posts. Candidates are the post on X, Search Engine Land's account and Business Insider's coverage. This page has not read their text and makes no claim about what they say.

What followed the heist?

According to Spring Agency, the site's organic traffic fell 42% after the founder wrote about the method on X. Spring presents the sequence as cause and effect: the post, then Google noticing, then the fall.

The supplied evidence cannot confirm that. It does not say whether Google took a manual action, changed how its systems treated the pages, or whether the fall had another cause, and it does not date the fall at all.

The 42% also has no stated baseline. A fall from a peak, a fall from an 18-month average and a fall from the pre-heist level are three different events with three different meanings.

Was the heist worth it?

It cannot be settled from the supplied evidence, because the one outcome that matters to a business, conversions, is described without any numbers. Spring reports that the founder says the extra traffic bought conversions, then points out that nobody knows what those conversions were.

Spring also raises costs the traffic figures do not capture. Its questions are how many visitors clicked an unhelpful page and left with a worse view of the company, and whether a short burst of conversions justified the cost if traffic later returned to its old level.

Those are open questions in Spring's account, not findings. The honest answer is that the traffic spike is reported, the long-run value is unknown, and anyone quoting a verdict in either direction is going beyond the record.

What does Google's guidance say about this method?

Google's guide says that creating separate content for every possible query variation, including fan-out queries, primarily to manipulate rankings violates its scaled content abuse spam policy. This is official guidance, read on 29 October 2026, and it describes today's rules.

That date matters. The evidence does not establish when each policy named in the guide was introduced, so the guide should not be read back onto 2023. What it does show is Google's present position on the shape of the heist.

Google's guide to generative AI features makes four relevant points.

• A high quantity of pages does not make a website higher quality or more relevant to users.
• Content that restates what is already available, or that a generative AI model could easily produce, adds little. Google calls this commodity content and contrasts it with content that has a unique point of view or first-hand experience.
• Anyone using generative AI tools to help create content must still meet Search Essentials and the spam policies.
• Google's AI systems can recognise a page's relevance even without an exact match to the query, so a page per phrasing is not needed.

A sitemap holds addresses, which is to say topics and phrasings. It holds no experience, data or point of view, so titles taken from it can only be filled with the common knowledge a model already has.

Was AI content the problem, or the scale?

On the evidence supplied, the problem was the purpose and scale of the publishing, not the use of AI. Spring says Google's February 2023 guidance put the focus on content quality, not on how the content is produced.

Google's current guide takes the same position from a different angle. It does not ban AI-assisted drafting, and it says such work must meet the same standards and spam policies as any other.

Spring offers a practical test of quality that is worth keeping. A model can produce predictive text that could be handed in the same form to anyone, and it can state out-of-date facts, as in its example of a model still naming a former UK Prime Minister as current. Those two failures, sameness and error, are what a human gate exists to catch.

What is documented and what is analysis?

The lists below separate what the supplied sources state from what Kanso infers from them. The documented list is attributed, and the analysis list is opinion and should be weighed as such.

• Documented (Spring Agency): a sitemap export, titles from URLs, about 1,800 AI articles, 3.6 million views in 18 months, 489,509 sessions in an October, and a 42% fall in organic traffic.
• Documented (Spring Agency): the founder says the traffic bought conversions, and the nature of those conversions is not given.
• Documented (Google, 29 October 2026): variation pages made to manipulate rankings violate the scaled content abuse policy, and quantity does not equal quality.

• Analysis (Kanso): a sitemap supplies topics and phrasings but no first-hand content, so the articles were likely commodity content.
• Analysis (Kanso): the 42% fall cannot be attributed to any single cause from the supplied evidence.
• Analysis (Kanso): the sums on this page mix units and are only a way to expose missing baselines.

Which six gates should AI-assisted pages pass before publishing?

Six gates, each phrased as a question, would have stopped the pattern this case describes. The table was built from the four Google points above and the questions Spring raises, plus one gate that is Kanso's own view.

The last gate is why Kanso keeps a person in the loop. Its Writer Agent turns Search Console opportunities into long-form drafts in your brand voice, for you to review and publish yourself. Kanso does not promise rankings.

The wider question of where approval belongs is covered in a risk-by-task matrix for AI marketing work. For a weekly habit that checks originality before you publish, see the weekly growth checklist.

What is still unverified?

Most of the case is unverified, and a reader should check six things before repeating any of it. This page was built from two supplied documents, so these gaps are the limits of the package, not proof that the facts are wrong.

• The dates of the sitemap export, the article publishing and the traffic peak.
• The unit and the analytics tool behind the 3.6 million figure and the 489,509 sessions.
• The baseline and date of the 42% fall.
• Whether Google took any action, and what kind.
• The identity of the competitor and its own traffic over the same period.
• What the conversions were.

For a traffic fall of your own, a seven-step diagnostic for small sites shows how to find the baseline and the cause before drawing a conclusion. For how Google's AI features fit with ordinary SEO, the GEO explainer separating official guidance from guesswork covers it.

The SEO heist: frequently asked questions

Sources

• The SEO Heist - A Cautionary Tale About AI, Spring (springagency.co.uk)
• Optimizing your website for generative AI features on Google Search, Google Search Central