Most editorial policies are a page of adjectives. Rigorous, unbiased, reader-first — words a site with something to hide would also use, because they cost nothing to write. So this page tries to avoid them entirely and describe mechanisms instead: who does what, in what order, with what they are allowed to know at the time, and what happens automatically when someone gets it wrong.
Where a control is weak, it says so. A team of 1 cannot run an independent audit function, and pretending otherwise would undermine the parts of this that are real.
How does a card review actually get made?
Nine stages, in this order, every time. The order matters more than any individual stage: the negative case is written before the positive one, and the score is set before anyone involved could know what the card is worth commercially.
Terms capture
Credit cards desk
The issuer’s own pricing-and-terms disclosure is read line by line and transcribed into a structured record: purchase and cash-advance rates, penalty terms, fee schedule, foreign transaction fee, balance-transfer mechanics, the welcome offer and its spend window. Nothing is taken from a marketing page, a press release or another publisher.
GateThe record is stored with the date it was reconciled and a link to the source document. A record with no source link does not save.
Earning and perk modeling
Credit cards desk
Every earning category and every named benefit is entered separately, with an annual value where one can be defended and a usability discount that says how much of it a normal cardholder will realistically capture. A statement credit that requires a specific booking channel is not worth its face value and is not recorded as if it were.
GateAny perk carrying a value must carry a discount and a one-line justification for it.
Redemption pass
Redemptions desk
The card’s currency is examined for what it can actually buy: which partners are reachable, which of them released bookable saver space in our availability log, and which sweet spots survive current pricing. A transfer list is not an asset until seats exist at the end of it.
GateAny redemption cited in a review must have been priced to a payment screen in the current quarter.
Valuation application
Data desk
Dollar figures — bonus value, realistic perk value, first-year value, net annual cost — are computed from the published valuation model. They are never typed by a writer, and a writer cannot override them. The model in force is Q3 2026.
GateThe build fails if a card figure appears anywhere in prose instead of being derived from the record.
Scoring
Editor-in-chief and credit cards desk
The card is scored on five axes against the published rubric, with a written rationale that publishes verbatim on the review. Scoring happens on the modeled record alone.
GateThe scorer has no access to revenue reporting, commission rates or network terms for any card. This is the firewall described below.
The negative case
Assigned writer
Before any positive copy is written, the writer must produce the “not for” list and the cons: who should not get this card, and what is genuinely bad about it. A review whose cons are cosmetic is sent back.
GateAt least one con must be a reason a reasonable person would decline the card outright.
Consumer-credit review
Nicholas Miles
Every review is read specifically for the reader who might carry a balance, be near an issuer’s application limit, or be about to take on a fee they cannot use. That reader gets a named paragraph in every review.
GateNo review publishes without this pass. It is the reason we say plainly that rewards never outrun interest.
Compliance and link build
Editor-in-chief
The advertiser disclosure, the “Terms Apply” mark, the rates-and-terms link and the issuer’s required disclosure text are attached structurally, by the components that display the offer, rather than by a human remembering to add them.
GateAn automated audit runs against the built site and fails the deploy on a missing disclosure, an uncloaked issuer link, or a compensated link without sponsored markup.
Publication and the review clock
Credit cards desk
The page publishes with a visible last-verified date and enters the 30-day re-verification cycle described below.
GateA record that passes its window is flagged in the editorial queue and blocks a clean build.
Who scores a card, and on what?
Scores are set by the editor-in-chief together with the credit cards desk, working from the modeled record: the terms as transcribed, the earning structure, the perk values after their usability discounts, and the redemption pass. They are not set by the writer of the review, and a writer cannot change one — a score moves only by being re-run through the rubric, with a written rationale that publishes on the card page verbatim.
Each axis is scored out of five, to one decimal. What each axis is asking, and what actually moves it:
| Axis | The question it asks | What raises it | What lowers it |
|---|---|---|---|
| Rewards value | What does ordinary spending on this card actually return, at our valuation of the currency it earns? | Broad categories that match real household spending; a currency with strong transfer options; earning that does not depend on a booking portal. | Rates that look high but sit on narrow categories; a currency with a hard cash floor and few partners; caps that bite before the year is out. |
| Perks value | How much of the benefit package will a normal cardholder capture, not how much is printed on it? | Credits that apply automatically or to spending you would do anyway; protections that pay out; lounge access with real coverage on your network. | Credits split into monthly slices, tied to one merchant, or requiring enrollment; benefits that duplicate something the reader already has. |
| Fee justification | Does the card still make sense in year two, once the welcome offer is gone? | A recurring benefit that covers most of the fee for a mainstream user; a fee that is honest about what it buys. | A card whose entire case is the first-year bonus; benefits whose realistic value falls below the fee for anyone but a frequent traveler. |
| Flexibility | How many good exits does the currency have, and how badly are you hurt if one closes? | Transfers to multiple unrelated programs; partners in more than one alliance; a currency that survives a single partner being dropped. | A single-airline currency; transfers only to partners that price like cash; a program with a history of overnight partner removals. |
| Overall | Would we recommend this card to a specific person we know, and who is that person? | A clear, nameable reader for whom this is the right first choice. | A card that is nobody’s best option — competent everywhere, correct nowhere. |
When does the commercial side enter the process?
At stage eight, and never before. Until the score is locked and the review is written, nobody working on the card has access to what it pays.
Concretely, that means the people who score cards do not have access to the revenue reporting system, do not receive per-card or per-issuer commission statements, and are not in the conversations where network terms are negotiated. This is a small company, so we will not pretend the firewall is architectural in the way it would be at a large publisher — it is a rule about who sees which report, enforced by the editor-in-chief, and it depends on people keeping it.
What is not merely a rule is everything downstream of the score. The rankings are computed. The dollar figures are derived. The disclosures are attached by the components that display an offer. Those are enforced by code that fails a build, and they are the reason the firewall is worth something even in a company this size.
Do issuers see our content before it publishes?
No. Not before publication, not in draft, not under embargo, not as a courtesy.
No issuer, affiliate network or agency has ever seen an Award Rate article before it published, been offered the opportunity to comment on a rating, or been given advance sight of a valuation change. We do not send drafts for factual review — where we need to confirm a fact we confirm it against the issuer’s own published disclosure, which is a document, not an opinion.
We do accept corrections from issuers after publication, on exactly the same footing as corrections from readers: send us the document, and if we got it wrong we will fix it and log it with the correction. An issuer has no faster route to a change than a reader does.
How are ranked lists ordered?
Ranked lists are the part of a rewards site most likely to be quietly sold, so ours are built to make that difficult. A “best cards” page does not contain a hand-written order. It declares a set of cards and a sort key, and the order is computed at build time from the card records and the valuation model.
The complete set of keys a list may sort on:
- Our rating
- Bonus value
- First-year value
- Lowest annual fee
- Lowest net cost
- Card name
Commission rate, network, and whether we are paid at all are not fields the sorting code can read. To move a card up a list, someone would have to change its published score or one of the numbers derived from its record — both of which are visible on the page, and both of which carry a rationale and a changelog entry. That is not a guarantee of good faith; it is a guarantee that bad faith leaves a trace.
Editorial judgement still decides which cards are eligible for a list and what the list is about. That judgement is exercised in the introduction of every ranked page, in writing, with the criteria stated before the cards appear.
How do we handle a conflict of interest?
The conflicts in this field are not exotic. Writers hold the cards they write about, know people at the issuers, and have loyalty balances that a devaluation would hurt. Pretending otherwise would be worse than disclosing it.
- Every member of the editorial team files a register of the cards, elite statuses and loyalty balances they personally hold, updated when it changes. Holding a card is not a bar to writing about it — the experience is most of what makes the writing worth reading — but it is disclosed on the review.
- Nobody may write about, score, or rank a product in which they hold a financial interest beyond being a customer: no equity, no consulting relationship, no paid speaking engagement with the issuer, no employment by the issuer or its agencies within the past two years.
- Personal referral links are banned outright. No one on this site may earn a referral bonus from a reader’s application, on any platform, at any time. All compensation arrives through the company, from the networks named in the advertiser disclosure.
- No one with knowledge of an unpublished valuation change may act on it — including transferring their own points ahead of publication. Valuation changes are embargoed internally until the quarterly release.
- Gifts, upgrades, free stays, press trips and hosted flights are not accepted. Where a route or property had to be experienced to write about it, we paid for it, and the review says how.
- A writer who develops a relationship with a source at an issuer discloses it to the editor-in-chief and is recused from scoring that issuer’s products.
How often do we re-verify an offer?
Card terms are reconciled against the issuer’s own pricing-and-terms disclosure on a rolling 30-day cycle, and within five business days of any change we become aware of. Every card page displays the date its terms were last reconciled, and that date is read from the record rather than typed, so it cannot be optimistic.
A record that passes its window does not quietly persist. It is flagged in the editorial queue, and the build reports it. Where a figure cannot be reconciled, it is suppressed rather than guessed: an empty field is a smaller failure than a confident wrong one.
Welcome offers get the same treatment with a harder consequence. If we cannot confirm an offer is live, the apply link for that card is switched off everywhere on the site from a single data field — 2 cards are in that state right now — and the card keeps its page, honestly labeled, instead of disappearing.
What would make us drop a card?
“Dropping” a card means different things depending on what went wrong, so the response is graded. The one thing we do not do is remove a page quietly, because a page that vanishes is indistinguishable from a page that was bought.
| Trigger | What happens | Timing |
|---|---|---|
| The issuer withdraws or pauses the offer | The apply link is switched off site-wide from one data field. The card stays visible with an honest “offer not currently available” state rather than sending a reader to a page that no longer shows what we described. | Immediate |
| We cannot reconcile the terms against the issuer’s own disclosure | Every derived figure for that card is suppressed until it reconciles. We publish nothing about a product whose terms we cannot source, however much traffic the page earns. | Immediate |
| A benefit we scored on is removed or materially cut | The perk record is corrected, the score is re-run through the rubric, and the change is logged. If the card falls out of a ranked list as a result, it falls out. | Same week |
| The issuer introduces a term we consider predatory | Deferred interest, a fee structure designed to be missed, or marketing that implies approval — any of these and the card is delisted entirely, with the reason published on the review. | Immediate |
| The card stops being a serious option for our readers | A card that has fallen behind its category is moved to comparison-only: it stays in the database so the comparison stays honest, and the call to action is removed. | Next review |
| An affiliate network asks us to change coverage | The request is declined and logged. If a network makes continued payment conditional on coverage, we end the relationship and keep the card, unpaid, with a label saying so. | Immediate |
How do we correct an error?
We do not silently edit a factual claim. When we get something wrong, the correction is published on the page where the error appeared and logged on one public corrections page, with what we said, what is actually true, how the error was found, and what changed in our process so it does not recur.
The previous claim stays visible. A corrections log that only shows the fixed version is a marketing page; the value of the log is that you can see what we believed before.
Material errors — anything a reader could have acted on financially, which in this field means ratios, expiry rules, fees and offer terms — are corrected within two business days of being confirmed, and acknowledged within one. Where an error touched a valuation, the currency is re-sampled rather than nudged.
What role does AI play in this site?
No page on this site is written, drafted or rewritten by a language model, and none carries a synthetic byline. If it has a name on it, that person wrote it.
Software does a great deal of work here, and it would be dishonest to imply otherwise. It collects award and cash prices, diffs an issuer’s terms page against the record we hold, ranks lists from published keys, generates the freshness dates, and refuses to complete a build if a card figure has been typed into prose or a disclosure is missing. That is automation of checking, not of judgement, and it is the reason a team of 1 can maintain 17 cards without the pages going stale.
Where a model is used internally — to flag a discrepancy between a terms document and our record for a human to read — its output is never published, and a human confirms the source document before anything changes.
How we treat sources, reader reports and unpublished data
Almost everything on this site is sourced from documents anyone can read: issuer pricing-and-terms disclosures, published award charts, program terms and conditions, and our own pricing observations. Where a claim rests on one of those, we link to it.
Reader-submitted data points — a transfer that posted faster than published, an award that priced differently from the chart — are welcome and are treated as leads, not as facts. A single report never becomes a published claim. It is reproduced by our own desk first, and if it cannot be reproduced we say that the report exists and that we could not confirm it.
We rarely rely on unnamed sources, and we do not publish an unattributed claim about a future devaluation or a coming offer at all. Rumour coverage is how a site in this category earns a week of traffic and a permanent discount to its credibility.
What to do if you think we got this wrong
Write to editors@awardrate.com. Everything to that address is read by a person, and complaints about our editorial judgement go to the editor-in-chief, who is named, has a bio, and is the person who owns the score you are arguing with.
If your complaint is that a figure is wrong, include the page and what you believe the correct figure is; that route is faster and it is the one that produces a correction. If your complaint is that our judgement is wrong, say so plainly — we publish the reasoning behind every score precisely so it can be argued with, and disagreements that change our mind end up in the changelog.
Frequently asked questions
Does an issuer paying you more commission move its card up a ranked list?
No, and the mechanism is worth understanding rather than taking on faith. Ranked lists on this site are not hand-ordered. They are sorted by one of a fixed set of published keys — Our rating, Bonus value, First-year value, Lowest annual fee, Lowest net cost or Card name — computed from the card record and the valuation model. Commission rate is not a field the sorting code can see. Changing an order therefore means changing a score or a number, both of which are published with a rationale and a changelog entry.Has an issuer ever asked you to change something?
Yes. Requests to soften a con, to remove a comparison, and to take down a paragraph about an application restriction have all been made, and all been declined. A request to change coverage is logged, and where a network has made payment conditional on it, we have kept the card as an unpaid listing instead. That is why unpaid listings exist on this site at all.Who reviews the reviewers?
Honestly: the gates do, because there is no second person. Award Rate is a single-editor site, and claiming peer review would be a lie dressed as a trust signal. What every review actually passes through is machine verification — figures reconciled against the issuer's own disclosure, cross-card comparisons re-checked at build time, compliance audited on the built page — plus a named paragraph in every review written for the reader who might carry a balance. The honest limit of that control is that judgement is not machine-checkable, which is precisely why the underlying data, the method and the corrections log are all public.What happens to a review when the offer changes?
Nothing has to be rewritten, which is the point of how the site is built. No card figure exists in prose anywhere on Award Rate; every number is rendered from one structured record per card. When an offer changes we edit that record, and every page that mentions the card updates in the same build. The build refuses to complete if a figure has been typed into a page instead. Cards were last reconciled on August 6, 2026.Do you accept sponsored posts or guest articles with links?
No, in both cases, and the answer has never been anything else. We publish no sponsored content, labeled or otherwise, accept no guest posts, sell no links, and run no partner-content program. Every article on the site is written by a member of staff whose bio is published here.Is any of this site written by AI?
Yes — AI tooling is used in drafting and in data work, and this site says so plainly. Every page publishes under a named human editor who reviews and owns it, and the honesty of the claims does not rest on who typed them: figures render from records reconciled against issuer disclosures, comparisons are re-checked at build time, and the build fails when prose and data disagree. The judgement, the scoring and the responsibility are human, and each page names the human responsible.