Nine Checks That Catch a Fabricated Tool Round-Up
Most "best tools" articles are never fact-checked. You can tell in about five minutes, and you do not need to be an expert in the category. These are the nine checks we now run before publishing anything, written in the order that catches the most for the least effort — each one catches a class of error the next check assumes away. Run them on this article too. They came out of reviewing 80 of our own posts and finding fabricated products, wrong prices and studies that never happened.
This is a method, not a takedown. Every failure named below is one we found in our own published work.
Key Takeaways
- Order matters more than thoroughness. Checking a product's price before checking whether the product exists wastes the effort — we found five recommended products that do not exist (audit of 80 posts, 2026-09-02/03, ANALYSIS).
- The single fastest check is opening the vendor's own domain. If a product name appears only inside round-ups and never on the vendor's own site, treat the list as unverified.
- A hedge attached to an exact number is a warning sign, not a caution. "Typically around $15" with no source is weaker than a bare figure, because it signals the writer knew and published anyway.
- Claimed testing without a dataset is not testing. If an article says it ran trials, the prompts, dates and results should ship with it.
- Count the items against the title. We published a "Top 5" containing six entries and a "10 Types" containing twenty-five.
1. Does the product exist?
Type the vendor's domain directly. Not a search result, not a review site — the company's own website. Then look for a page describing the exact product name the article uses.
We found two invented product names attached to real companies, one recommended vendor whose domain was a parked page, and one that resolved to a domain-for-sale listing. In each case a single visit to the vendor's site would have settled it.
The signal: a product name that exists only inside other people's round-ups and never on a vendor's own site.
2. Is it even the right kind of product?
A real product in the wrong category is as misleading as a fake one, and it is easier to miss because everything checks out until you ask what the thing actually does.
We ranked a transcription platform first in a list of note-taking apps. We reviewed a fitness wearable as a to-do list app, complete with deadlines and team assignment it has never had. Both companies are real, well known, and take thirty seconds to check.
The check: read the vendor's own one-line description of itself, then ask whether it belongs in this article.
3. Is the price right?
Open the pricing page. Copy the plan names exactly as the vendor writes them. Note whether the figure is monthly or annual. Read the free-tier limits rather than the word "free".
About half the checkable prices in our round-ups were wrong, and nearly all of them made the product look cheaper than it is. Two products were credited with free plans that do not exist.
The check: if the price will not load, say so in the article. Substituting a figure from a comparison site is how stale prices spread.
4. Does the feature exist as described?
Open the product and confirm each listed capability. Vendors publish what they do; the information is usually right there.
We marked one tool down for having "no API for catalogue automation" when it sells an API with its own documentation — and that supposed gap was the stated reason for ranking it below a competitor. We described a company's tattoo product as scannable when its own page says, in italics, that it is not.
The signal: a criticism with no source. Negative claims need checking as much as positive ones, and they get checked far less often.
5. Do the links go where the anchor text says?
Resolve every outbound link, follow the redirects, and read the anchor text against the destination.
We found a link anchored "predictive analytics tools" pointing at an article about AI image generators. We found an auto-linker that had swallowed a sentence boundary — "saved per SKU. As AI becomes…" became a link to a Norwegian municipal property company. We found a link anchored "social media presence" that pointed at a service selling followers, where nothing in the visible text hinted at the destination.
That last one is the important pattern. A generic anchor concealing a commercial destination cannot be caught by reading the article. Only following the link finds it.
6. Is anyone being paid?
Grep the article's links for tracking parameters: ref=, via=, utm_, fpr=, gclid, partner IDs, and redirects through affiliate networks.
We found four buttons carrying another publisher's affiliate parameters, apparently copied wholesale from a competitor's round-up. We earned nothing from them and carried the disclosure obligation anyway. We also found a Google Ads landing URL with a live click identifier pasted straight out of an advertisement.
The check: if a link carries a tracking parameter, the article needs a disclosure. If it carries someone else's tracking parameter, something has been copied.
7. Can each statistic be traced?
Every number needs a named, dated, fetchable source. "Reports say", "studies show" and "experts agree" are not sources.
We found a consultancy's speed finding — code written 35–45% faster — restated as an automation finding: "automates 45% of coding tasks." Different claims, and the difference matters to anyone planning headcount against it. We found market-size figures restated in the present tense a year after their base year, from a research firm that had already replaced them.
The check: open the source and find the sentence. If the article's number is not in it, the number does not survive.
8. Does the article agree with itself?
Count the items. Count them against the title, the introduction and the conclusion.
We published a "Top 5" containing six entries, a "10 Types" containing twenty-five with one section duplicated, a title promising eleven tools over an introduction promising ten, and a comparison table listing fourteen apps where the body reviewed thirteen — with one of them priced but never actually reviewed.
Why this matters beyond tidiness: an article that cannot count its own sections was not read end to end before publishing. Everything else in this list becomes less likely.
9. Is the platform documentation current?
Anything describing someone else's interface, report name, metric or specification gets checked against that platform's current documentation on the day of publication.
We published a guide to an analytics report that had been renamed, using a navigation path that no longer existed, recommending an attribution model the platform removed nearly three years ago, and instructing readers to use a product retired in 2023. We cited advertising character limits to a help page that contains no character limits.
The check: platform docs are free and authoritative. There is no excuse for this class of error, and it is the one readers notice fastest because they are following along.
Two claims that should stop a draft outright
These are not "add a citation" problems.
Claimed testing with no dataset. We published "we spent 30 hours testing 12 AI photo generators" and "we ran 12 identical trials for each tool", with stopwatch timings to the second — and no prompts, no tool list, no date range, no results table. Only four of the twelve tools were even named. If a test was run, publish the dataset. If it was not, do not describe one.
Guarantees about regulated outcomes. We published that a tool "ensures regulatory compliance" and "keeps the company safe from fines". No software delivers a regulatory outcome. On health, financial or legal topics this is the most dangerous sentence pattern there is, and it is common.
Running these on someone else's article
You do not need to do all nine. Take three products from any round-up and run checks 1, 3 and 5: does the vendor's own site have a page for that exact product name, does the published price match, and do the links go where the anchor text claims.
If any of the three fails on more than one product, the article was not verified, and the rest of it should be read on that basis.
How This Guide Was Sourced
Written and maintained by the LogicBalls editorial team (logicballs.com). Disclosure: LogicBalls builds AI writing tools. Every failure named in this article is one we found in our own published content.
Method. These nine checks were derived from reviewing 80 published posts claim by claim on 2–3 September 2026, selected by search impressions. The order is the order we found most efficient in practice, not a theoretical hierarchy: each check catches a class of error the following checks would otherwise waste effort on.
They are now binding on us. The nine checks are written into our house style as a required pre-publish gate. Whether that is enforced by the writer or automatically is a separate question we have not finished answering.
AI involvement. The posts containing the failures described here were AI-assisted and published without verification. This post was AI-assisted and reviewed before publication.
No LogicBalls telemetry is used in this guide.
Frequently Asked Questions
How long does this actually take?
For a ten-tool round-up, roughly an hour if the vendors publish clearly and longer if they do not. Most of it is checks 1 and 3, which are mechanical. The order is what keeps it to an hour — checking a price before checking existence is wasted work.
Which check catches the most?
Check 1, by a wide margin, because it is binary and fast. Check 3 catches the most individual errors. Check 5 catches the ones that most embarrass a publisher, because a link to somewhere unexpected is visible to any reader who clicks.
Can this be automated?
Partly. Tracking parameters, count mismatches, hedge-plus-number patterns and link resolution are all greppable. Existence and category checks need a person to look at a page and think. The automatable parts are worth scripting; they are not the parts that catch the worst errors.
Are you saying most round-ups are fabricated?
No. We are saying ours contained fabricated entries, wrong prices and invented tests, and that nothing about how they were produced was unusual. Draw your own conclusion about how common that is, using these checks rather than our opinion.
Conclusion
Nine checks, in an order that matters: existence, category, price, capability, links, tracking, statistics, arithmetic, platform documentation.
They are not sophisticated. Every one of them is "open the thing and look". What makes them work is doing them in order and doing them before publishing rather than after a reader complains — and being willing to write "we could not verify this" when the answer is that you could not.
