How We Verify Pricing

Every number on PricingLog is supposed to earn your trust the same way: it was captured by a person, on a stated date, with saved evidence behind it. This page explains exactly how that works — and, just as importantly, what this site does not claim.

How prices are captured

I check each vendor's own pricing page directly, rendered from a US location, in a clean browser profile with no account logged in. Many vendors show different prices or currencies to visitors from different countries, so every price on this site follows one convention: US pricing, in USD, excluding tax. Both monthly and annual billing views are captured, because the gap between them is one of the most common sources of confusion in third-party pricing articles.

For every capture I save a full-page screenshot and a copy of the page HTML, filed by tool and date. Those files are the evidence behind each number. If a vendor appears to be showing different prices to different visitors (A/B testing), I note that on the page rather than presenting one variant as the only truth.

What "Last verified" means

Data pages carry a visible "Last verified" date. That date changes only when I have actually re-checked the numbers against the vendor's live pricing page — it is never bumped automatically to look fresh. If a date looks old, that is honest information too: it tells you exactly how stale the data might be. Prices can change at any moment between checks; always confirm on the vendor's site before paying.

How costs are normalized

Vendors price in incompatible units — words, credits, scans, seats — which makes sticker prices almost useless for comparison. Where this site shows a normalized cost, it uses these exact formulas:

• Cost per 100,000 words = plan price ÷ included words per month × 100,000, rounded to the cent. Credits are first converted to words using the vendor's own published ratio.
• Annual-effective monthly price = annual plan total ÷ 12, rounded to the cent. This is what a month actually costs when you pay yearly.

If an input is unknown, the cell shows a dash — never a guessed number and never $0. Plans without public pricing ("contact sales") are labelled as exactly that.

Two rules govern the conversion itself, and they are the reason some tools carry a figure and others do not:

• Native word allowances are used as published. Where a vendor states an allowance directly in words per month, no conversion happens at all.
• Credits are converted only with the vendor's own stated ratio. If a vendor publishes credits but never says what a credit is worth, that is where the calculation stops. A ratio is never back-derived by dividing a plan's price or volume by another plan's — that would manufacture a vendor definition out of arithmetic and present it as theirs.

When a tool cannot be normalized

Not every market prices on one model, and the AI detector category clearly does not. Forcing every tool into a single cost column would mean inventing numbers for the ones that do not fit, so instead each tool carries its pricing model as a label, and the cost column shows that label rather than a figure. A label is never rendered as a blank, a dash or $0, and non-comparable rows are grouped below the sorted ones rather than interleaved — a missing number must never sort to the top as though it were the cheapest option.

• Metered volume — the vendor publishes an included monthly volume and, where credits are used, a words-per-credit definition. These are the only rows that produce a computed cost per 100,000 words.
• Flat rate, per check — a fixed fee with a per-submission size cap and no monthly allowance at all. A per-check cap limits one scan; it is not a monthly volume and is never treated as one, so no per-word cost exists. It is shown as its own constraint column.
• Free only — no paid tier for the capability. The free terms are published as the vendor states them, and $0 never enters a ranking of paid plans.
• Institutional / contact-sales only — no public consumer price. Such a tool is only described that way when there is positive evidence of how the product is actually bought; the mere absence of a price from a page proves nothing and is not enough.
• Unit definition searched for and not found — the vendor publishes a volume in its own units but no definition of those units, and a bounded search of its public pages on the capture date did not find one. The pricing row still publishes, with the cost figure omitted. This records only that the listed sources did not contain a definition on that date; it is not a claim that none exists anywhere, and it never licenses deriving one.
• Annual-only allowances — some plans are billed yearly only and state their allowance per year rather than per month. Neither the price nor the allowance is divided by twelve to force it into the monthly formula: the plan is shown with its published annual figures and no standard cost per 100,000 words, because that number is defined on a monthly basis and inventing a monthly split would misstate what the vendor sells.

Where a page compares a mixed category, it states plainly how many tools carry a computed figure and how many do not. That mix is a finding about the market, not a shortfall to hide.

Promotional prices are not list prices. Where a vendor shows a limited-time or code-gated discount beside its standard price, the standard price is what enters the data and the comparison; the promotion is recorded as a dated observation only. A discount that expires would otherwise silently become a permanent claim.

Pricing history and the Wayback Machine

Historical prices come from two sources: my own dated captures going forward, and the Internet Archive's Wayback Machine going backward. Wayback snapshots are used only when the archived page actually shows rendered prices — many archived copies of modern pricing pages are empty shells, and those are not evidence. Every history entry links or refers to its source. History is append-only: past entries are never rewritten. If an entry turns out to be wrong, a correction entry is added; the record of the mistake stays.

Update cadence

Tracked pricing pages are checked on a weekly cycle, with a full re-check of each tool at least monthly. Detected changes are logged in the changelog with the old value, the new value, the date, and evidence. Weeks with no detected changes are logged too — a null result is still a result, and it proves the checks happened.

Corrections

If you find an error — a wrong price, an outdated limit, a missed change — please report it. I verify reports against the vendor's live page and aim to fix confirmed errors within 48 hours, with the correction noted openly in the changelog. Vendors are welcome to use the same process; corrections require evidence, not just a request, and accurate historical records are not removed on request.

Neutrality and money

Some links on this site may become affiliate links over time — the disclosure page lists exactly which programs are active (at launch: none). The rule that governs all of it: no commercial relationship ever changes a number, a table, or an ordering. Tables sort by data — price, cost per unit, verification date — never by commission. Tools with no affiliate program get identical treatment.

What this site does NOT claim

• No reviews. PricingLog does not publish "reviews" or experience claims about how good a tool is. Pages document what tools cost and what plans include — facts I can verify — not experiences I haven't had.
• No accuracy or quality testing. For tools where quality is contested (for example, AI detectors), this site makes no claims about accuracy. Where accuracy matters to a buying decision, I link to independent third-party studies rather than inventing my own verdicts.
• No ratings. There are no stars, scores out of ten, or "editor's choice" badges here, and no rating markup for search engines. A pricing record has no business rating things.
• No affiliation with any vendor. All product names and trademarks belong to their owners; they are used only to identify the products being discussed.

Revision history of this methodology

• 6 July 2026 — First published.
• 1 September 2026 — Added how credit-to-word conversion is bounded to the vendor's own stated ratio, how tools that cannot be normalized are handled (flat-rate, free-only, institutional-quote, unpublished unit definitions, annual-only allowances), and how promotional prices are treated. Published alongside the AI detector comparison.