Skip to content
Worix
BrowsePublish
Log inSign Up

task-product-seo-rewrite

Multi-step TASK — rewriting the SEO titles and meta descriptions of the merchant's products — drafts the merchant reviews on a card and saves to Shopify from it. Enter when: the merchant asks to write, rewrite, fix or improve the SEO titles or meta descriptions of their products ("rewrite my product SEO titles and descriptions", "write meta descriptions for my best sellers", "fix the SEO titles on my products"); the merchant chose products on the rewrite card, or pressed Rewrite on the product SEO check card (a hidden note that opens "The merchant chose products for the rewrite" or "Rewrite these"). Do NOT enter: checking products without changing anything (that is task-product-seo-check); rewriting products a product SEO check card on screen lists, asked in chat — they tick them on that card and press Rewrite, so point them to it in one sentence; a product's own description text, images, alt text, barcode, brand or category — you cannot change those, so say in one sentence that they edit them in Shopify; a blog article or social post (that is task-content-publish); greetings or small talk. Read this skill's full body BEFORE your first action — it carries the node graph you must report progress against, one node at a time.

7 downloads
Free
Reviewed
internal
task
marketing
requires-yepai-tools

Generated from tasks/product-seo-rewrite/task.yaml by /task-authoring.

Do not edit here — edit the definition and regenerate.

name: task-product-seo-rewrite description: >- Multi-step TASK — rewriting the SEO titles and meta descriptions of the merchant's products — drafts the merchant reviews on a card and saves to Shopify from it. Enter when: the merchant asks to write, rewrite, fix or improve the SEO titles or meta descriptions of their products ("rewrite my product SEO titles and descriptions", "write meta descriptions for my best sellers", "fix the SEO titles on my products"); the merchant chose products on the rewrite card, or pressed Rewrite on the product SEO check card (a hidden note that opens "The merchant chose products for the rewrite" or "Rewrite these"). Do NOT enter: checking products without changing anything (that is task-product-seo-check); rewriting products a product SEO check card on screen lists, asked in chat — they tick them on that card and press Rewrite, so point them to it in one sentence; a product's own description text, images, alt text, barcode, brand or category — you cannot change those, so say in one sentence that they edit them in Shopify; a blog article or social post (that is task-content-publish); greetings or small talk. Read this skill's full body BEFORE your first action — it carries the node graph you must report progress against, one node at a time. kind: task task_name: "Product SEO rewrite" label_template: "<products> SEO rewrite" tags: [internal, task, marketing, requires-yepai-tools] license: MIT-0 nodes:

  • id: scope label: "Choosing products" purpose: "Put the rewrite card up so the merchant picks which products to rewrite and, when product editing is not allowed yet, allows it first. The card stores their choice and builds the batch on the server." advance_by: user expects:
    • "the products to rewrite, chosen on the card" tools:
    • yepairag_marketing__SHOW_SEO_REWRITE_SCOPE next:
    • to: draft when: "the merchant chose the products — the rewrite card's or the check card's hidden note says so"
  • id: draft label: "Writing titles and descriptions" purpose: "Read the batch and write a new SEO title and meta description for each field it marks to fix, then put them on the review card. The server checks every draft; a refused one is fixed and shown again." advance_by: agent tools:
    • yepairag_marketing__GET_SEO_REWRITE_BATCH
    • yepairag_marketing__SHOW_SEO_REWRITES next:
    • to: review when: "SHOW_SEO_REWRITES accepted the drafts and the review card is on screen"
    • to: scope when: "the merchant asks to rewrite a different set of products"
  • id: review label: "Waiting for your review" purpose: "The merchant reviews each draft on the card, edits, approves or skips it, and presses Save; the card saves to Shopify. The card ends the task." advance_by: user tools: []

Task — Product SEO rewrite

Flow

flowchart TD
  scope("Choosing products") -->|the merchant chose the products| draft["Writing titles and descriptions"]
  draft -->|SHOW_SEO_REWRITES accepted the drafts| review("Waiting for your review")
  draft -.->|a different set of products| scope
  classDef terminal stroke-width:3px
  class review terminal

Rounded = waiting on the merchant · square = you are working · dashed = going back · bold border = terminal. The check card's Rewrite note opens the run at draft.

scope

Choosing products — Put the rewrite card up so the merchant picks which products to rewrite and, when product editing is not allowed yet, allows it first. The card stores their choice and builds the batch on the server.

  • The card is the only way products get chosen. Report scope, call SHOW_SEO_REWRITE_SCOPE and end the turn with no text: the card is the whole reply. This holds when the merchant already named the products in their message ("my best sellers", "the Summer collection") — they pick that on the card, which stores it on the server; you have no other way to hand products to the rewrite. Never call GET_SEO_REWRITE_BATCH before a card's note arrives.
  • Product editing not allowed yet (write_access: false): the card itself shows the Allow banner first and keeps Start disabled until it is granted. Say nothing about it and never ask for it in prose. Asked why it is needed, answer in one sentence: saving the new titles and descriptions needs permission to edit products, and nothing is saved until they press Save.
  • If the tool answers duplicate: true, the card is already up — say nothing.

draft

Writing titles and descriptions — Read the batch and write a new SEO title and meta description for each field it marks to fix, then put them on the review card. The server checks every draft; a refused one is fixed and shown again.

  • A card's note is the go-ahead, and the products are already chosen. Two notes start this node: "The merchant chose products for the rewrite: <label>…" from the rewrite card, and "Rewrite these N products…" from the check card's Rewrite button. The second one opens the run here — report draft and never put the rewrite card up. Then call GET_SEO_REWRITE_BATCH (no arguments).
  • A batch with no rows means none of those products needs a new title or description: say so in one sentence, from its message, and stop — no card.
  • Write exactly what the batch marks to fix — what the merchant asks for does not widen it. Per row: a seo_title when fix.title is true, a meta_description when fix.meta is true — every field marked true gets a draft. A field marked false never goes in the call, even when you would word it differently and even when the merchant asked for both fields or for products that already pass, because the server refuses it: either it is fine as it is, or the row lists it under blocked (blocked: {meta: "no_description"} — the product has no description to build one on). The card itself says which chosen products it left out and why, and which fields already pass, so none of that is yours to say. Asked after the card to rewrite more than the batch holds ("both fields for all eleven", "the ones that are already fine too"), your whole answer is ONE sentence — this rewrite only covers what the SEO check flagged, and the card lists the rest — with no tool call. reasons says what is wrong with the current text.
  • Every fact comes from the row. Materials, sizes, colours, fit, compatibility, origin, care and certifications come only from what the row carries — its title, product type, vendor, tags, options (sizes, colours), collections and description text. A fact the row does not state is left out — never filled with a plausible guess. Never a review, rating, award or claim the row does not make. Use a Brand IQ keywords term only when it truly describes this product, and at most once.
  • Never count characters. The server measures every title with the shop_name_suffix the theme adds after it, and every description, and refuses any that does not fit. Write short and let it measure.
  • Show the drafts, then report review, then stop. One SHOW_SEO_REWRITES call with your drafts, each row referred to by its row number. When it answers status: "shown" (or duplicate: true — the card is already up), the card is on screen and the task has reached its waiting node: your next call is the review report, and it is the turn's last call. The turn has no text — not before the card, not between the card and the report, not after it; the card is the reply. Stopping at the card without that report leaves the merchant's progress stuck on "Writing titles and descriptions".
  • A refused call is fixed silently. The error lists every problem, row by row. Fix exactly those — a description refused for carrying none of the row's facts, or because it "reads like" another row, is rebuilt around that product's own facts, not reworded — keep every other draft as it was, and call again with the full set — no text to the merchant, no narration of the fix. A field refused three times is left out of the next call: the server accepts any subset of rows, and any subset of a row's fields marked true.
  • A change asked after the card ("make the Hydrogen one shorter", "don't mention shipping" — never a field or product the batch leaves out, which stays out as above): report draft, change only what they asked, call SHOW_SEO_REWRITES again with every row (the others exactly as before), then report review — no text.
  • Name products by their title, never by an id, in the rare sentence you write here.

SEO title

From coreyhaines31/marketingskills skills/seo-audit/references/title-tags.md (MIT, see NOTICE), adapted for product pages:

The title is the strongest on-page relevance signal you control and the first thing a searcher reads, so it has two jobs. It has to match the query, and it has to win the click against nine other results.

[Product type + key attribute] + [what makes this product worth clicking]
  • Query first. Put it in the first few words, where both the ranking and the reader's scan start. For a product, the query is what a shopper types: the product type and its key attribute ("all-mountain snowboard", "linen apron").
  • Then the reason to click: an attribute the row states — a material, a style, what it is for ("for filming runs").
  • No shop name, no brand. The theme appends the row's shop_name_suffix ( – <shop name>) to every title, so the shop name written into the title shows twice — and the server refuses a title that contains it.
  • Google truncates by pixel width, not characters, so keep the important words first.
  • The product's own name may lead when it says what the product is ("Hydrogen Snowboard | Science-Inspired Graphic").

Only use "free," a year, or a number when the page delivers on it. A "2026" title on a page last updated in 2024 is a promise the content breaks. Here the page is the row: no "free", year or number it does not state.

Keep the set consistent:

  • Title, H1, URL slug, and meta description should all describe the same topic. They don't need identical wording, but a reader moving from the title to the page should never wonder whether they landed in the right place. On a product page the H1 is the product's own title.
  • Every title is unique. Duplicates split relevance and usually mean a template is filling in the same string everywhere. The server checks the batch and the rest of the store.
  • Don't change slugs. Change the title; leave the URL and its links alone. A product's handle is never changed, suggested or mentioned.

Google replaces the title with its own text when a title is:

  • Much longer than the display width
  • Stuffed with keywords or repeated boilerplate ("| Brand | Brand Blog | Home")
  • Out of line with the page's H1 or content
  • Generic ("Home," "Products," "Untitled")

Remove words that add length without meaning ("Ultimate," "Best," "Premium").

Common fixes:

ProblemBeforeAfter
Says nothingUntitled product / SB-2041All-Mountain Snowboard with Pixel Cloud Graphic
BloatedThe Videographer Snowboard – Best Snowboard For Filming Your Runs In Any ConditionsVideographer Snowboard | Board Built for Filming Runs
StuffedSnowboard | Snowboards | Best Snowboard | Snowboard SaleFreestyle Snowboard with Teal Abstract Graphic
Shop name in the titleLiquid Snowboard | <shop name>Liquid Snowboard | Blue Landscape Graphic
Duplicatetwo products both Snowboard | <shop name>each names its own product: Hydrogen Snowboard | Science-Inspired Graphic
Promise the page breaksFree Shipping Ski Wax 2026 (the row says neither)All-Temperature Ski and Snowboard Wax

The review card is the owner's line-by-line table — current text, your proposal and the reason, approved or skipped one product at a time. One proposal per field keeps the review fast: never offer alternatives.

Meta description

From coreyhaines31/marketingskills skills/seo-audit/SKILL.md, "Meta Descriptions" (MIT, see NOTICE), adapted for product pages:

Check for:

  • Unique descriptions per page
  • Includes primary keyword
  • Clear value proposition
  • Call to action — only one tied to this product's own facts ("Pick the 154 or 158 cm length"), never a bare "Shop now" or "View the product details"

Common issues:

  • Duplicate descriptions
  • Auto-generated garbage
  • Too long/short
  • No compelling reason to click

Two short sentences about THIS product: each carries at least one fact that is true of this product and not of the others in the store, taken from the row — its type, a tag, an option such as its sizes or colours, a collection, or something its description says. The banned form is the sentence that would fit any product in the store ("Explore this snowboard for winter sport and snowboarding. View the product details before you choose your board.") — and a batch whose descriptions share one frame with the nouns swapped. Before the call, read each description against its row: if it would still be true with another product's name on it, it has no fact yet. The server refuses both forms — a description with none of the row's facts, and a row that reads like another row. The server measures the length. Never repeat the title — and never open with the product's name or title: a description that starts with it is flagged by the check and refused by the server, so open with what the product is, does or looks like. Never add the store or brand name, and never put in a price, a discount, a currency, a percentage or any number that changes (stock, sale dates) — it goes stale, and the server refuses it.

Search, AI search and what never to do

From coreyhaines31/marketingskills skills/ai-seo/SKILL.md (MIT, see NOTICE):

Google's position (AI features optimization guide):

"The best practices for SEO continue to be relevant because our generative AI features on Google Search are rooted in our core Search ranking and quality systems."

Google explicitly says:

  • No special markup or files are required for AI Overviews or AI Mode
  • Don't chunk content for AI — write for people, organize with normal headings and paragraphs
  • Don't write separate content for AI — that risks "scaled content abuse" spam policy
  • Helpful, reliable, people-first content wins — same E-E-A-T standards as regular Search

When in doubt, write for people and organize for clarity; that satisfies both camps.

What NOT to do:

  • Write separate content "for AI". Same content should serve people and AI. Writing variants targeted at AI systems risks the scaled content abuse spam policy — Google's words.
  • Generate at scale for ranking manipulation. AI-generated content is fine if it meets Search Essentials and spam policies. Mass-producing thin variations does not. Every product gets its own words, never one sentence with the name swapped.
  • Pursue inauthentic mentions. Don't fabricate citations, invent awards, run "independent" review sites you own, or bulk-spam Reddit/Wikipedia for AI visibility. Fake or misleadingly sourced reviews can also break the FTC's 2024 reviews rule. Real participation only.
  • Keyword stuffing — Unlike traditional SEO where it's just ineffective, keyword stuffing actively reduces AI visibility by 10% (Princeton GEO study)

review

Waiting for your review — The merchant reviews each draft on the card, edits, approves or skips it, and presses Save; the card saves to Shopify. The card ends the task.

  • After Save. The card saves to Shopify itself and then sends "Saved N products" with a hidden note that opens the same way. The task ended when the card went up, so that message gets no node report and no tool call. Answer it with ONE short sentence that confirms the save with the note's number — and, when the note has "Not saved" lines, that those were not saved and the card says why — no product list, and no promise about rankings or traffic. Then stop.
  • Asked when the change shows on Google: it can take a few days. Asked to undo one: they change it in Shopify on that product's page, under its search engine listing — you have no tool to undo a save.

How a task runs

This part is the same for every task; only the nodes differ.

1 — Entering the task. Report the first node before doing anything else. That report is what opens the run: it tells the backend a task has started and gives the merchant their first progress line.

2 — Entering each node. Report it the moment you enter, before doing that node's work. One report per node. A single turn usually passes through several nodes with no input from the merchant in between — reporting as you go is what lets them watch the work happen instead of staring at nothing until you finish. A node whose work is a single tool call still gets its report first.

3 — Reaching the end. A terminal node is one with no way out except back to itself. Reporting it completes the task — the backend works that out from the graph, so there is nothing extra for you to send.

4 — After completion. A completed task is not sealed. If the merchant later asks to change something, report the node that owns it again and the task is simply unfinished again. Never open a second task for a change to work you already did.

How to report

yepairag_task__REPORT_TASK_STEP
  task: "product-seo-rewrite"
  node: "<the node you just entered>"
  • Two fields, both read straight off the graph above. The step's label, its number, the total and whether the task just completed are filled in by the backend — you cannot get them wrong because you do not send them.
  • The reply carries legal_next: the only nodes you may report next.
  • ok: false means nothing was recorded and the run has not moved. Correct the node id and report again rather than carrying on as if it had been accepted.
  • Never restate any of it to the merchant. It drives a progress bar they are already looking at.

Download

ZIP package — ready to use

Skill Info

Creator
Yepai
License
MIT-0
Downloads
7
Published
Nov 12, 2026
Updated
Nov 20, 2026