How the Listing Grader scores — and why the generator gives a different number
Por Oleksandr Shaniuk · fundador de Sellura · Actualizado el 4 de septiembre de 2026
Sellura shows you two SEO numbers, and they are not the same number computed twice. The free Listing Grader scores a listing that already exists, from its public page. The generator scores a listing Sellura just wrote from your inputs. Different weights, different inputs — the same listing can honestly get a 71 from one and an 88 from the other. This page prints both rulers in full so you can check our work instead of trusting a badge.
Sellura is a multi-marketplace listing tool — Etsy, eBay, Amazon Handmade, Shopify and Depop — and is not affiliated with any of them. Nothing below describes how those marketplaces rank listings internally. It describes what our own code measures.
Two scorers, and the difference is what each one can see
The grader fetches a public listing page, so it is limited to what a marketplace shows the world: the title and the description. It cannot see your tags — on Etsy a seller's 13 tags are hidden from shoppers entirely — so it doesn't score them. Guessing would demo better, and would be fiction. The generator scores a listing it just produced from what you typed in, so it sees everything: title, tags, description, materials, attributes. Five factors instead of four.
That is the whole reason the numbers differ. Neither one is stricter or better calibrated. The grader is scoring a smaller listing, because a public page is a smaller listing.
The free grader: four factors, 100 points
The weights are fixed and sum to 100: keyword front-loaded 30, title length 25, title keywords 20, description 25.
Front-loading is deliberately blunt here. Nobody typed a product name into the grader, so it checks only whether the first 40 characters contain a substantive word — one that survives a stopword list and a fluff list ("gift", "cute", "perfect", "unique", "beautiful"). Pass and you get all 30; fail and the factor fails outright rather than nudging you. A title opening "Perfect Gift For Her, Cute Unique Idea…" has said nothing in the space that matters most.
Title length is cap-aware. On the keyword-school marketplaces credit rises toward the recommended length, but past the platform's hard cap it is halved and the advice inverts: instead of "use the space", the grader tells you to trim it yourself and keep the wording you chose. That exists because of a real defect — Amazon cut its title cap from 200 characters to 75 on 27 July 2026, and until the fix a live 150-character Amazon title scored full marks for length and was told to grow.
The description factor adds two things: up to 15 points for length, scaled against 300 characters, and a further 10 for an opening that names the product rather than greeting the shopper. That second half is the one people lose without noticing — a warm description opening "Thank you so much for visiting my shop!" scores zero on it, because a description's first line is indexed text and that one spent it on nothing.
The generated-listing score: five factors, and tags are the biggest
The generator's weights are different: keyword front-loaded 20, title relevance and length 20, tags 30, description 20, materials and attributes 10.
Tags at 30 is the largest single factor anywhere in Sellura, and the one the free grader cannot score at all. Half of it is completeness — using every slot the marketplace gives you — and half is quality, measured as the share of your tags that are multi-word phrases. Thirteen single-word tags fill every Etsy slot and still earn only half the factor, because single words compete against the entire marketplace.
The fixed constants are the lines you are measured against: the front-load window is the first 40 characters, a description counts as full-length at 300, and the attributes factor wants 3 materials and 3 input attributes. Below those it scales proportionally — three of six is half, not a failure.
The score is rule-based on purpose, and the AI never computes it
Both scorers are deterministic. The same listing always produces the same number, and every factor traces to a line of code with a threshold in it. There is no model in the scoring path. AI writes the listing text and phrases the fixes — but it sits strictly on top of the score and never inside it, which is the part that matters: the engine that writes your listing is graded by an instrument it cannot influence. A model marking its own homework will always tell you it did well.
This costs us something. A model-judged score would be more flexible with listings our rules read badly. It would also be unauditable: it could return 84 today and 79 tomorrow, and neither you nor we could say why. A number you can't reproduce isn't a measurement. We would rather ship a ruler that is occasionally wrong in a way you can point at — which, as the next section admits, has happened.
What both share: thresholds, and the marketplace's title school
The pass lines are identical in both. A factor passes at 75% of its own maximum and warns at 45%; overall, a listing passes at 75 and warns at 50. A 74 and a 76 are not meaningfully different listings — they sit on either side of a line we chose.
Both also branch on the marketplace, and not just by swapping character limits. Each one carries a title style. Etsy is "natural": its Seller Handbook guidance on listing titles, published August 2025 and updated since, moved away from long keyword chains toward short scannable ones — clearly state the item, put the most important traits up front, "consider using less than 15 words", try not to repeat words, and move subjective and gifting phrases out of the title. So on Etsy the length credit is a plateau, not a climb: a maxed-out 140-character title scores worse than a natural 80-character one, not better.
Everywhere else the older rule still holds — eBay, Amazon Handmade, Shopify and Depop are "keywords" marketplaces, where using the title budget genuinely helps and the score rewards it. Etsy's hard cap is 140 characters against a target of 80; eBay 80 and 75; Amazon Handmade 75 and 70; Shopify 70 and 60. Depop is the odd one out and worth stating plainly: it has no title field and publishes no character count at all, so both of its numbers are ours — we score the first line of the description against a 65-character target because that is roughly one phone-width line, not because Depop said so. One deliberate softness: over-long Etsy titles taper gently rather than falling off a cliff, because a great many listings were written to the older advice and an abrupt drop would read as a penalty. Etsy doesn't publish how its search treats those titles, so we don't claim to know either.
Etsy title rules, in depth → · Free title checker — the same rules, no account →
An asymmetry we haven't closed: Amazon's second headline box
Amazon Handmade has two indexed headline fields, not one: the title, and a 125-character Item Highlights field that Amazon indexes alongside it. The generator knows this and scores both against a combined budget, so a flawless 70-character title with an empty highlights box does not get full marks — 125 searchable characters are going unused, and the score says so.
The free grader cannot do this. It parses a public product page, and that field is not something it can reliably read back off one. So on Amazon Handmade the grader is scoring roughly two-thirds of the indexed headline while reporting a number as if that were the whole thing. It will read more optimistically than the generator, and the gap is exactly the field it cannot see. We would rather write that down than let the two numbers quietly disagree.
The time we found our own ruler was wrong and fixed the ruler
In September 2026 we re-scored 111 real generations and found the largest single source of lost points was our own front-load factor. Listings we had just written — good ones — were being told "Keyword front-loaded 13/20 · Improve".
The bug: the factor demanded the first three words of the seller's product field, literally, inside the first 40 characters of the title. Against real listings that meant "parks" didn't match "Park" and "striped" didn't match "Stripe". Filler the writer had rightly dropped — "commemorative", "famous", "retro-style" — cost a third of the factor each. Worst of all, a product named adjective-first, like "Sobriety Recovery Commemorative Coin", had its actual item noun in fourth position, where the check never looked. A factor named "is the keyword front-loaded" was really measuring "did you repeat the seller's words".
The first fix made matching tolerant of word forms and added an item-noun check — and over-corrected. It still gave full credit only when two-thirds of the seller's words appeared, so on a padded product line like "Sobriety Recovery Commemorative Coin – 2 Year" the echo scored 20 out of 20 while "2 Year Sobriety Coin…" — the better title, the one a buyer would type — scored 15. The second fix changed what the factor asks: a buyer types the item plus a qualifier or two, never the seller's every word, so the ruler now wants the item noun plus at least one descriptor.
The part worth taking away is what we did not do. At no point did we lower a pass threshold or reweight a factor to make the numbers look better. When a good listing scores badly, the first suspect is the ruler. We now re-score the same real listings before and after any change to the scorer or the writing prompt, precisely so that a "fix" which merely inflates scores is visible as one.
What the score does not promise
It is not a prediction of your search position. Neither scorer models a marketplace's ranking algorithm, because none of them publish one. What the score measures is whether your listing follows the rules those marketplaces do publish, plus the search conventions that apply to any indexed text.
We have no marketplace API access: no live search-volume data, no competitor rank data, no view or impression counts for your shop. When another tool shows you a search-volume figure, it is worth asking where it came from. We don't show one because we don't have one.
And the grader only sees what is public — not your tags, attributes, photos, price, shipping, reviews or shop history, several of which matter more to a sale than any of this. A listing scoring 100 has every signal we can measure working in its favour: a title that leads with the item as a buyer would type it, tag slots filled with long-tail phrases rather than single words, a description that opens on the keyword, attributes complete, all of it inside that marketplace's own limits. That is the half of ranking you control, written the way the platform asks for it — which is not the same as ranking, let alone selling.
These rules are also our reading of published guidance at a point in time, and guidance moves — Etsy rewrote its title advice and has revised it since, and Amazon cut its title cap by more than half in July 2026. When it moves, this page and the code move with it.
- Not a ranking prediction — no marketplace publishes its algorithm, and we don't model one.
- No API access: no search volume, no competitor data, no view counts.
- The grader is blind to tags, attributes, photos, price and reviews.
- A 100 means every signal we can measure is working for you — it is not a ranking prediction, and not a sales forecast.
The two rulers, side by side (points out of 100)
| What is measured | Free Listing Grader | Generated-listing score |
|---|---|---|
| Keyword front-loaded in the first 40 characters | Free Listing Grader30 | Generated-listing score20 |
| Title length / title relevance and length | Free Listing Grader25 (length only) | Generated-listing score20 (relevance + length) |
| Title keyword variety, or natural-title stuffing check | Free Listing Grader20 | Generated-listing scorefolded into relevance |
| Tags | Free Listing Gradernot scored — not public | Generated-listing score30 |
| Description | Free Listing Grader25 | Generated-listing score20 |
| Materials and attributes | Free Listing Gradernot scored — not public | Generated-listing score10 |
| Factor pass / warn thresholds | Free Listing Grader75% / 45% of the factor | Generated-listing score75% / 45% of the factor |
| Overall pass / warn | Free Listing Grader75 / 50 | Generated-listing score75 / 50 |
FAQ
Why does the same listing get two different scores in Sellura?
Because two different scorers ran, over different amounts of information. The free grader reads a public listing page, so it scores only the title and description across four factors. The generator scores a listing built from your own inputs, so it also scores tags, materials and attributes across five factors — with tags alone worth 30 of the 100 points. The gap is mostly the tags the grader could never see, not a disagreement about your listing.
Is the SEO score generated by AI?
No — the score is measured, not judged. The same listing always produces the same number, and every factor traces to a threshold published on this page. AI writes your listing text and phrases the fixes; it never computes the score. That separation is the whole design: a model that graded its own output could return 84 today and 79 tomorrow for the same listing, and neither you nor we could say why. The ruler itself is calibrated against real listings — when we re-scored 111 of them and found good titles being marked down, we rebuilt the ruler rather than move the number.
Does a score of 100 mean my listing will rank on the first page?
No, and we won't imply otherwise. Neither scorer models a marketplace's ranking algorithm, because none of them publish one. What a 100 does mean is that every signal in your listing text is working: the item named up front the way a buyer searches it, a title length judged against that marketplace's own title school rather than one generic rule, every tag slot carrying a long-tail phrase, a keyword-led description, attributes complete. Ranking also depends on price, photos, reviews, shipping, shop history and competition — none of which any text scorer can see, ours included.
Why doesn't the free grader score my tags?
Because it cannot see them. The grader works from a listing's public page, and tags aren't public on every marketplace — Etsy hides a seller's 13 tags from shoppers entirely. Scoring them would mean inventing them, so instead the grader states that it graded the public title and description only. To get tags scored, generate a listing in Sellura, where the tags are ones you actually have.
Two scores, two rulers, and the difference between them is coverage rather than strictness: the grader sees a public title and description, the generator sees the whole listing including the tags that carry 30 of its 100 points. Both are rule-based and reproducible, and both bend to the marketplace, because Etsy's short natural title and eBay's packed keyword title cannot be judged by one rule. Both also stop short of the promise that sells best — neither predicts where you rank, because nobody outside those companies can. What a good score does mean is that the checkable part of your listing is done properly, and that the words are no longer the reason nobody found you.