This page describes how Fog & Clarity articles are made: where the facts come from, how claims are worded, what we refuse to publish, how AI is used, and what happens when we get something wrong. If a page on this site does not follow these rules, that is an error and we want to hear about it.
Every factual claim traces to a primary source
We treat the following as evidence: peer-reviewed papers, linked to their PubMed record or DOI; Cochrane reviews; NIH Office of Dietary Supplements fact sheets; FDA and FTC documents, including warning letters and Federal Register notices; court records; and regulatory registers.
We never use the following as evidence: supplement vendor blogs, affiliate review sites, brand-hosted “study” pages, and press-release newswires such as accessnewswire, newswire.com, barchart press releases and their syndication into finance portals. That last category matters more than it sounds: a press release dressed as a news article is the standard way a claim with no trial behind it acquires the appearance of a source. We cite such a page only as evidence about marketing, never as evidence that something works.
If a claim cannot be traced to an acceptable source, we cut the claim. We do not fabricate citations, trial names, PubMed IDs, dates, doses or quotes. A fabricated citation is worse than a missing one: it is a lie the reader can catch. Every source in a Sources list is one we opened, and every number in the body appears in the paper it is attributed to — as in our account of the two large ginkgo prevention trials, where the sample sizes and the null results are the story.
Every effect claim names its population
This is the single rule that separates a defensible page from an indefensible one. “Improves memory” is false about every compound we cover. What can be true is “improved delayed recall in adults over 65 with mild cognitive impairment,” or “reduced reaction time in healthy young adults an hour after a single 200 mg dose.”
So we report who the effect was found in, at what dose, for how long, and how large it was — in numbers rather than adjectives. A standardized mean difference of 0.29 is a small effect and we say so. We do not generalize from a deficient population to a replete one, from an impaired population to a healthy one, or from older adults to younger ones. Those are different questions, and here they routinely have different answers. The tier definitions and worked examples are on How We Rate the Evidence; the ratings themselves are collected in our tier list.
Mechanism is labelled as mechanism, anecdote as anecdote
A plausible mechanism is not an effect. Most marketing in this category runs on the gap between the two: a compound raises nerve growth factor in cell culture, or NAD+ in blood, and the reader is invited to finish the sentence. When we describe preclinical work we say that is what it is, and we say whether the human trial that would test it has been run.
Community reports are similar. They can be informative about side effects, and they are not trial data. Where we include them they sit in a labelled box, never mixed into a paragraph about study results.
Where nobody has run the study, we say so. “No evidence exists” is a publishable answer here, and it is often the most useful sentence on the page.
The standard we hold memory claims to
In January 2017 the FTC and the New York Attorney General sued the marketer of Prevagen over memory and cognition claims made for apoaequorin. In December 2024, after a jury trial, a federal court ordered the marketer to stop making the deceptive memory and cognitive-improvement claims. The full history is in our page on the apoaequorin verdict.
That ruling is the benchmark against which we word any memory claim here. In practice: we never write that a compound restores or reverses a number of years of memory; we never describe a supplement as improving memory without naming the population and the trial; and we never repeat a vendor’s claim on the assumption that liability stays with the vendor. It does not transfer. If this site says it, this site has to be able to support it.
We also never write that any ingredient treats, cures, prevents, reverses or mitigates a disease. Not dementia, not Alzheimer’s, not mild cognitive impairment, not ADHD, not depression.
What we will not recommend
We never recommend vinpocetine, the racetams, noopept/omberacetam or phenibut, and we do not link to products containing them. FDA’s position is that these are not lawful US dietary ingredients, and it has issued warning letters on that basis. Explaining why to avoid them is exactly the kind of article we want to publish; recommending them is not. Huperzine A is discuss, never recommend — it is a reversible acetylcholinesterase inhibitor, the same class as donepezil.
How we use AI
We use AI tools as research assistants — to find candidate papers, to summarize long documents, to draft structure, and to check our own text for the failure modes this site cares about, such as an effect claim that has lost its population.
We do not use AI as a source. A language model is not evidence, and it is capable of producing a citation that looks perfect and does not exist, which is precisely the failure this site cannot afford. So the rule is simple: every source is opened and read by a person, and every number is checked against the document it came from, before publication. A human editor is responsible for every page, decides every rating, and is accountable for every claim on it. No page is published straight out of a model.
Review cadence
Evidence moves, and regulatory positions move faster: NMN’s US legal status reversed in September 2025. So pages are re-checked on a schedule rather than left to rot.
- Quarterly — pages that turn on a live regulatory question, and any page naming a specific product, its price or its formulation.
- Every 12 months — every evidence and dosage page, checked against new trials, updated meta-analyses and revised Cochrane reviews.
- Immediately — whenever a major trial, safety signal or enforcement action lands, or a reader reports an error.
A review that changes nothing does not change the page. A review that changes a conclusion changes the page and leaves a dated note.
Medical review
No article here is medically reviewed unless a named reviewer appears on it. At present, none do. We would rather say that plainly than display a “medically reviewed” badge with no name attached, which is common in this category and means nothing. When we add reviewers, their names, credentials and the review date will appear on the page.
Corrections
When we find an error — or a reader finds one for us — we fix the page and add a dated note explaining what changed. We do not silently edit a claim out of existence. Typographical fixes are made without a note; anything that changes a claim, a number, a dose or a conclusion gets one. If a claim turns out to be unsupportable, it comes down and the note says so.
To report an error, email hello@fogandclarity.com or use the contact page, with the URL and — if you have it — the source you think we should have used.
Commercial independence
A small number of pages carry affiliate links and we earn a commission when a reader buys through one. No page whose subject is safety carries a commercial link, and we take no payment for a review or a ranking position. Details are on the Affiliate Disclosure page.
These statements have not been evaluated by the Food and Drug Administration. Nothing here is intended to diagnose, treat, cure or prevent any disease.
This article is information, not medical advice. Talk to a doctor or pharmacist about your own situation, especially if you take prescription medication.
