The 13 Checks: How Geopad Scores Your Products

Why We Publish Our Scoring Methodology

When a tool gives your product a 61/100, you are entitled to ask two questions: what exactly was measured, and why should the thresholds be believed? Most tools in this category answer neither. The score arrives as an authority you are meant to accept, and the methodology stays a trade secret - usually because there is less behind it than the interface implies.

We think the score is only useful if you can argue with it. So this post documents every one of the 13 checks Geopad runs on each product, the exact weight each carries, the thresholds it applies, and - the part you will not find elsewhere - an honest account of which thresholds rest on published evidence, which reflect how search engines mechanically behave, and which are our own working heuristics that we expect to refine as real data accumulates.

One thing to state up front: this is a catalogue quality score, not a ranking prediction. It measures whether your product data is complete, structured, and machine-readable. It does not measure your backlinks, your reviews, your domain authority, or your prices against competitors - and a large share of what determines search and AI recommendation outcomes lives in exactly those places, outside the catalogue. The score tells you whether your side of the street is clean. That is all it claims.

How the Maths Works

Each check is a pure function: it reads a product's data and returns a score between 0 and 1, plus any issues it found. Issues carry one of three severities - critical (something is missing entirely), warn (present but weak), or info (an improvement opportunity).

Each check also carries a fixed weight. The weights across all 13 checks sum to 110, and your product score is simply the weighted sum normalised to 0-100. Nothing adaptive, nothing personalised, no hidden factors - two products with identical data always get identical scores, and re-running a scan on unchanged data returns the same number.

The weights are deliberately skewed. The four heaviest checks - SEO title (15), meta description (15), description depth (12), and image alt text (12) - account for 54 of the 110 points, roughly half the score. That reflects a judgement: these are the fields search engines and AI systems read first, and the fields most often left empty or duplicated on real stores. The remaining nine checks contribute at the margins, which is where they belong.

The Big Four (54 of 110 points)

1. SEO title - weight 15. Missing entirely: critical, score zero. Present, we check length against a 30-60 character window - under 30 costs more than over 60, because a too-short title wastes the space search engines give you, while a too-long one merely truncates. We also apply a small penalty when a product has no dedicated SEO title and falls back to the product name: the fallback works, but a tailored title is written for the search result, not the product page. The 30-60 window is convention-backed: it derives from how search engines truncate display titles at roughly 600 pixels. It is not a magic number, and a 62-character title is not a crisis.

2. Meta description - weight 15. Missing: critical. Under 70 characters: half credit - you are leaving the pitch space empty. Over 160: modest penalty, because search engines truncate the rest. The 70-160 window is, again, display-truncation convention. Meta descriptions do not directly influence rankings; they influence whether a human clicks your result, which is the point.

3. Description depth - weight 12. Under 20 words: critical - at that length, search engines and AI assistants have essentially nothing to read, and an empty description is flagged in its own words rather than as "0 words," which reads like a bug. Under 80 words: a warning. These two cut-offs are honest heuristics - there is no published study establishing that 80 words is the line where a description becomes useful. What is well-established is the direction: thin content gives extraction systems nothing to extract. Treat 80 as a floor we chose, not a law we found.

4. Image alt text - weight 12. No featured image at all: critical, score zero - this is the one check where the penalty is really about commerce basics rather than SEO. Otherwise the score is simply the fraction of images that carry alt text, with the severity escalating when more than half are missing. Alt text is one of the least ambiguous items on this list: it serves accessibility, image search, and machine description of your products simultaneously, and writing it is purely mechanical work.

Structure and Extraction (32 of 110 points)

5. Headings - weight 8. A binary check: does the description HTML contain any H1-H3 heading? Unstructured walls of text are harder for both readers and machines to parse. Fails soft - 40% credit rather than zero - because plenty of short, adequate descriptions legitimately need no headings.

6. Description structure - weight 6. Where check 5 asks whether any structure exists, this one measures how much: three or more H2 sections earns full credit, one or two earns partial, formatting without sections (bold text, lists) earns less, and plain unstructured prose scores lowest. The reasoning is mechanical: extraction systems lift answers from clearly bounded sections far more reliably than from continuous prose. The three-section threshold is our heuristic; the underlying mechanism is not.

7. Specifications table - weight 4. Full credit for an actual HTML table; partial credit when specification-type facts (material, dimensions, weight, capacity and similar) appear in prose; lowest when neither exists. A table is the most literally machine-readable format a product page can contain. This is the lightest-weighted check on the list, deliberately - many product types have no meaningful specs to tabulate.

8. Structured data readiness - weight 8. An honest caveat first: without crawling your storefront, no app can prove what JSON-LD your theme actually renders. So this check scores whether the fields needed to generate complete Product schema - vendor, product type, price - are present in your product data, and starts from partial credit rather than zero. It measures readiness, not deployment.

9. Internal links - weight 6. Does the description link to at least one other page on your own store? Internal links give crawlers paths between your products and collections. Info-level severity, half credit when absent - useful, rarely urgent.

Hygiene (24 of 110 points)

10. Duplicates - weight 8. If another product in your store shares the same SEO title, or the same URL handle, both facts are flagged as critical and each costs half this check's score. This is the least negotiable check on the list: duplicate titles force search engines to guess which page to show, and the guess is frequently wrong. This check runs across your whole catalogue in one pass, which is why it exists in an audit tool rather than a per-product checklist.

11. URL handle - weight 5. Missing: critical. All-numeric ("/products/48291"): heavy penalty - the handle is a ranking-relevant field and numbers describe nothing. Over 75 characters: mild penalty. Structural, not controversial.

12. Freshness - weight 5. Products untouched for over a year score 40% on this check; over six months, 80%. We will be straight about this one: the specific decay curve is our heuristic, and "freshness" is the check most often over-claimed in this industry. Updating a product does signal ongoing maintenance to crawlers; it does not magically improve rankings. Five points of weight is what that reality deserves.

13. Speed proxy - weight 6. Products carrying more than 8 images get flagged, because oversized galleries are a common cause of slow product pages. Note the word proxy: this is not a Core Web Vitals measurement - it is a data-level indicator of the most common self-inflicted speed problem in product catalogues. If you need real performance measurement, use PageSpeed Insights; this check exists to catch the 40-image gallery before it becomes a mystery.

What the Score Deliberately Does Not Measure

Being clear about the boundary matters more than the checks themselves.

The score does not measure off-site authority - reviews, backlinks, press, marketplace presence. Independent research on AI shopping recommendations consistently finds that third-party sources carry substantial weight in what gets recommended, and none of that lives in your catalogue. A 95-scoring product on an unknown store can still lose to a 60-scoring product with five hundred reviews.

It does not measure keyword strategy. We check that your title and description exist and are well-formed; we do not score whether you chose the right keywords for your market, because that is a judgement about your business, not a property of your data.

It does not predict rankings or citations. Catalogue quality is the necessary condition - the work that makes everything downstream possible - not a sufficient one. Any tool telling you a data-quality score predicts your traffic is selling you the correlation it hopes you will not question.

And it does not run on a schedule by itself. You trigger scans and optimisation runs when you choose to; every published fix goes through your approval or your explicitly enabled auto-apply setting.

Frequently Asked Questions

Why do the weights sum to 110 rather than 100? Because the two newest checks (description structure and specifications table) were added to the original eleven with their own weights rather than by rebalancing everything - the score normalises to 0-100 regardless. Re-weighting the whole table is on our list; silently changing every merchant's historical scores to make the arithmetic prettier is not.

Will my score change if I change nothing? Only through the freshness check, which decays slowly past six and twelve months. Every other check is deterministic on your data.

Is 100 the goal? No. Several checks encode trade-offs - a legitimately spec-free product will never max the specifications check, and does not need to. Treat the score as a prioritised to-do list, not a grade. The severity labels (critical first) are the actual work order.

Do these checks matter for AI search as well as Google? The structural checks - sections, tables, alt text, schema readiness - exist precisely because extraction systems, including AI assistants, parse bounded structure far more reliably than prose. But we will not promise citations: AI-referred traffic remains a small fraction of e-commerce sessions today, and the honest reason to do this work is that it improves how every system, human and machine, reads your catalogue.