How do you handle dead retailer pages?

A separate dead-link checker re-resolves every outbound URL on a rolling 24-hour cycle. Deals whose retailer page returns a 404, 410, or unreachable response are flagged dead_at and removed from every public listing. The deal page itself returns a 410 Gone so search engines drop it from the index quickly.

How We Score Deals — Methodology and Editorial Standards

Q: What does the deal score actually mean?

Our deal score is a 0–100 composite of three weighted signals: discount magnitude vs. the 90-day baseline (50%), product-match confidence (30%), and price-tier suitability (20%). Anything under 50 is filtered out before it can rank.

Q: Why a 90-day baseline instead of MSRP?

MSRP is a number the manufacturer prints on a sticker; it has no relationship to the price the retailer was actually charging last month. A 90-day price history captures the rolling median of what the item really costs day-to-day, so a 25%-off sticker is only meaningful if the price is below that median.

Q: Do you remove deals when the price comes back up?

Yes. Crawls run hourly and a deal whose current price no longer beats its 90-day baseline drops out of the ranking automatically. The history stays — you can still load /deal/{id} and see the chart — but the deal won't surface on the homepage or category pages.

Q: Are AI tools involved in scoring?

ML is used for parsing (title extraction, image detection, product matching against our catalog) and category classification. Scoring itself is rules-based and deterministic — given the same price history, every deal gets the same score every time. We chose deterministic scoring over an ML score so the rankings are auditable and a regression is easy to debug.

"On sale" is a marketing word, not a meaningful one. Most retailers can sticker any price as a discount by raising the strikethrough number until the math works out. That's why every deal we list goes through the same four-stage gauntlet before it can rank — so a 25%-off sticker only shows up here when it's actually 25% off the price the retailer was charging last month, not 25% off a fictional list.

The four checks, in order

1. Crawl freshness

Deals enter our pipeline through a continuous crawl of major US deal feeds (DealNews, Slickdeals, retailer-specific RSS). Each item gets a timestamp at the moment we observed the price. Anything older than an hour falls out of the homepage ranking automatically — a stale price is the single biggest source of "I clicked through and the price was different" complaints, so we just don't show it.

2. Outbound link health

A separate worker re-resolves every outbound URL on a rolling 24-hour cycle. If the retailer page now returns a 404, 410, redirects to a category page (the seller's "dead-product fallback"), or fails to load at all, the deal is flagged dead_at and disappears from every public listing. The deal detail page itself returns HTTP 410 Gone so search engines drop the URL from their index quickly.

3. Discount integrity

This is the one most affiliate sites skip. Every deal is matched (where possible) to a 90-day price history — both our own crawl record and, for Amazon items, what we can pull from the Amazon Product Advertising API and external trackers. The current price has to beat the rolling 90-day median, not just the strikethrough on the page. That filter alone cuts about a third of the raw inventory we ingest.

For items where 90-day data isn't available (a new product, a one-off retailer feed), we fall back to comparing the discount-percentage claim against the typical fluctuation range for that retailer. A "20% off" claim from a retailer whose normal weekly variance is 15% gets discounted accordingly.

4. Score threshold

What survives all three previous checks gets a deterministic score, 0 to 100, computed as:

score = 0.5 × discount_norm
      + 0.3 × match_score
      + 0.2 × price_norm

Where:

discount_norm — the discount percentage clamped to 0–100 with a soft cap at 70% (anything claiming more is almost always a data error or fake list price)
match_score — confidence (0–100) that we correctly matched the listing to a known SKU, used to deflate scores for ambiguous matches that could be a knockoff or a different generation
price_norm — a small bias toward more accessible price points (a $30 deal at 40% off ranks ahead of a $3,000 deal at 40% off, all else equal — the broader audience benefit is real and worth weighting)

The minimum score floor is 50. Anything below that is hidden from the homepage and category pages, regardless of how flashy the headline number looks.

The freshness decay function

Once a deal passes the four checks, where it ranks within the homepage feed is governed by a Hacker-News-style freshness decay so today's 70-score beats yesterday's 90-score:

effective_score = deal_score / (hours_since_enriched + 2)^1.5

That gravity exponent (1.5) is tuned so a 95-score deal that's a day old falls below a fresh 70-score, but a fresh 95 still wins handily. The result is a homepage that turns over daily without dropping high-quality finds in the first hour.

What we don't do

We don't pay for placement. Retailers cannot buy a top spot. The score formula is the same for every deal regardless of source.
We don't accept editorial influence. No retailer or PR firm has any input on which deals are featured or how they're written about.
We don't fake aggregate ratings. Some affiliate sites surface fabricated star-counts to game rich snippets. We do not. Every rating-style indicator on this site is sourced from real click-through data or omitted.
We don't republish dead deals. Once dead_at is set, the deal stays gone — we don't resurrect old URLs to inflate inventory counts.

Why this transparency matters

Most affiliate "deal" sites are black boxes. You can't see how they decide what to feature, you can't audit their freshness claims, and you can't tell whether a "best of" list is editorial or sponsored. That opacity is the entire reason people stopped trusting deal aggregation.

Publishing the methodology in full doesn't help our ranking algorithms; arguably it lets competitors copy them. We do it anyway because if you're going to trust the deals on this site, you should be able to verify how they got there. If you find a problem with the math, email [email protected] — we will fix it.

Frequently asked questions

What does the deal score actually mean?

A 0–100 composite of discount magnitude (50%), product-match confidence (30%), and price-tier suitability (20%). The minimum publishing floor is 50.

Why a 90-day baseline instead of MSRP?

MSRP is a number on a sticker, not a price the retailer ever charged. The 90-day rolling median captures what the item actually costs day-to-day; a real discount has to beat that, not the strikethrough.

Do you remove deals when the price comes back up?

Yes — automatically, on the next hourly crawl. The deal disappears from rankings; the history page stays accessible so you can audit what happened.

Are AI tools involved in scoring?

ML handles parsing and classification. Scoring itself is deterministic — same inputs, same score, every time. We chose this so rankings are auditable.

How AIFindsDeal Scores a Deal