We Audited Our Own 530 Blog Posts. Five Recommended Products That Do Not Exist.
We checked every product recommendation on this blog against the vendor's own website. Five of them were for products that do not exist. Two were invented product names attached to real companies. One was a parked domain with no business behind it. One was a domain listed for sale. One was a fitness tracker reviewed as a to-do list app, with deadlines and team assignment it has never had. All five were published, indexed, and recommending themselves to readers.
This post is the method and the findings. It covers 530 published posts, 80 of which went through full claim-by-claim review on 2 and 3 September 2026. We are publishing it because a company that sells AI writing tools should be the first to show where AI-assisted writing fails, and because the check that found these takes about ninety seconds per product.
Key Takeaways
Five recommended products could not be shown to exist. Two were invented product names, one resolved to a parked GoDaddy domain, one to a domain-for-sale listing, and one was a real company in the wrong category entirely (audit of 80 posts, 2026-09-02/03, ANALYSIS).
The failure is not hallucinated text. It is an unverified list. In at least two cases the name appears to have been lifted from a third-party round-up where a publisher's name was read as a product name.
The check is trivial. Open the vendor's own domain and confirm the product has a page. Not a search result, not a review site — the vendor's own site (CodeAnt AI, retrieved 2026-09-03, SOURCED).
Scoring systems do not catch this. The posts containing fabricated products scored 62–77 on a 100-point content rubric. A rubric checks whether a citation is present, not whether the subject is real.
Nothing here was deleted. Every correction was published in place, naming the earlier wording, so readers can see what changed.
What we checked, and how
The method matters more than the findings, because the findings are only useful if you can repeat them.
We pulled all 530 published posts to local files, then reviewed 80 of them ranked by search impressions. For every named product in those posts we did four things, in this order:
Resolve the vendor's own domain. Not a search result and not a review site — the company's own website.
Find a product page for the exact name used in our post. If our post said "X Compliance Suite", we looked for a page describing a product called X Compliance Suite.
Confirm the category. Does the product do the thing the article is about?
Re-read every price and capability claim against the vendor's current pages, with the retrieval date recorded.
Anything that failed step 1 or 2 is what this post is about. Steps 3 and 4 produced a separate set of findings, covered in our post on pricing errors.
Where a vendor's site blocked automated requests, we recorded that as unverified rather than guessing. That happened often enough to be worth saying: several vendors return HTTP 403 to anything that is not a browser.
The five
1. A compliance product that was never sold
A post on healthcare compliance platforms recommended a "Compliance Suite" from an AI company, describing how it tracks compliance tasks, monitors data access and alerts staff to risk.
The company is real. The product is not. Its own homepage describes something entirely different: "AI agents that reason across your code, infrastructure & runtime to prove what is exploitable and fix it" (CodeAnt AI, retrieved 2026-09-03, SOURCED). It is a code-security platform. It sells nothing to healthcare compliance officers.
ANALYSIS — how this probably happened. That company publishes a blog post titled "Top 11 HIPAA Compliance Software Options". Our article appears to have taken the publisher's name off a round-up and turned it into one of the products in the list. The article was about HIPAA — a health-data regulation — which makes it the worst possible place for an invented recommendation.
2. A file-transfer product that was never sold
The same pattern, in a different post. An article recommended a "HIPAA Tracker" from a managed file-transfer company.
The company is real and genuinely useful for moving protected health data. It describes itself as "Managed SFTP/FTPS Cloud Storage as a Service" (SFTP To Go, retrieved 2026-09-03, SOURCED). There is no product called HIPAA Tracker, and nothing on the site tracks compliance.
3. A recommended vendor that was a parked domain
An article about SEO reseller services recommended a company, described its white-label packages, its local SEO offering and its monthly reporting.
The domain is a parked GoDaddy page. The root redirects to a /lander path served from domain-parking infrastructure, with an empty page body, no product, no pricing and no contact details (retrieved 2026-09-03, ANALYSIS). There is no business there. There may never have been.
The description in our post was three specific sentences about services this company offered. It could not have been verified by anyone, because there was nothing to verify.
4. A recommended platform that was a domain-for-sale listing
An article on AI tarot platforms reviewed six products. One of them, described with daily card-draw reminders and a messenger-style interface, resolves to a domain-for-sale page. No matching product could be found anywhere (retrieved 2026-09-03, ANALYSIS).
5. A fitness tracker sold as a to-do list app
The strangest one. An article on the best to-do list apps reviewed a company that makes a fitness and health wearable — a wrist-worn device and a membership subscription. Our post described its deadlines, its team assignment and its task management.
It has none of those. It is not a task manager and has never been one.
This is a different failure from the first four: the company is real, well known, and easy to check. Nobody checked.
Why a quality score does not catch this
Every one of these posts had been scored against a 100-point content rubric covering content quality, SEO, expertise signals, technical elements and citation readiness. They scored between 62 and 77.
A rubric checks whether a citation is present. It cannot check whether the subject is real. A post recommending a product that does not exist can be well structured, correctly formatted, internally linked, and score in the seventies — right up until a reader clicks through and finds a parking page.
That is the gap. Automated content scoring measures the shape of an article. It does not open the links.
The check that catches all five
Ninety seconds per product, and it would have caught every case above:
Type the vendor's domain directly. Not a search — the domain.
Find the page for the exact product name you are about to publish. If you cannot find a page for that name, do not publish that name.
Read what the company says it does, in its own words, on its own homepage.
Ask whether that matches the article you are writing. A code-security platform in a healthcare compliance round-up is a category error even when the company is real.
If a product name only appears in other people's round-ups and never on the vendor's own site, that is the signal. All four of the invented or misplaced names in this post share that property.
What we did about it
Nothing was deleted. Every affected post was corrected in place, and each correction names the earlier wording so a returning reader can see exactly what changed. Where a claim could not be verified in either direction, it is labelled as unverified rather than quietly removed.
We also wrote the check into our own house style as a required step before anything is published. It is now the first of nine, ordered so that each catches something the next one assumes away — there is no point pricing a product that turns out not to exist.
How This Guide Was Sourced
Written and maintained by the LogicBalls editorial team (logicballs.com).
Disclosure: LogicBalls builds AI writing tools. This post is about failures in our own published content.
Method. 530 published posts were pulled to local files. 80 were reviewed claim by claim on 2–3 September 2026, selected by search impressions. Every named product was checked against the vendor's own domain, in the order described above. Vendor claims are quoted from the vendor's own pages with the retrieval date attached; findings derived from our own review are marked ANALYSIS.
What is verifiable here and what is not. The vendor descriptions are quoted and linked — you can check them. The parked-domain and domain-for-sale findings are stated as of 2026-09-03; domains change hands, and a reader checking later may see something different. The count of five is specific to the 80 posts reviewed, not to all 530.
AI involvement. The posts that contained these errors were AI-assisted and published without source verification. This post was AI-assisted and verified against the sources linked above before publication.
No LogicBalls telemetry is used in this guide.
Frequently Asked Questions
Does this mean AI wrote fake products into your blog?
Not exactly, and the distinction matters. In at least two cases the name came from a real third-party round-up where a publisher's name sat next to product names. The failure was that nobody opened the vendor's site before publishing. A model producing an unverified list is doing what it does; publishing it unchecked is an editorial decision.
Why publish this instead of quietly fixing it?
Because a quiet fix leaves every reader who acted on the old version with no way to know. And because we sell AI writing tools — an argument about verified output is worth very little from a company unwilling to show its own failures.
How do I run this check on someone else's round-up?
Take any three products from it. Type each vendor's domain directly and look for a page describing that exact product. If a name only exists inside round-ups and never on a vendor's own site, treat the whole list as unverified.
Were the other 450 posts checked?
No. 80 posts were reviewed in full. The remaining posts have had structural and link-level checks but not claim-by-claim verification. We are not going to describe them as verified when they are not.
Conclusion
Five recommended products across 80 posts could not be shown to exist. The check that finds them is opening the vendor's own website, and it takes about ninety seconds.
The more useful finding is what did not catch them: a 100-point quality rubric, internal linking checks, and structural review all passed these posts. Scoring measures the shape of an article. Only opening the links tells you whether the thing you are recommending is real.
