SEO
Engine
Finds the listings search engines cannot rank, drafts the missing texts with Copilot and proves the result with the figures from your Google Search Console – before and after. No drafted listing text reaches a channel before a person has applied it.
What does the SEO Engine do in Business Central?
The SEO Engine audits the marketplace and shop listings in the merchantCENTRAL hub against rules you set – every listing except those of blocked items or variants and those of channels where Audit SEO Content is switched off. It checks title, meta title, meta description, description, URL slug, images and identity data such as brand and EAN. Every breach becomes a finding with a severity and one sentence on what search engines make of it. Where something is missing, Copilot can draft it from the facts of your item – as a proposal that a person applies or rejects. The audit itself needs neither AI nor any external service.
A test bench for every listing
The audit judges what the channel actually serves, not what happens to be stored somewhere – and every audited listing gets a score that tells you whether the work is paying off.
Findings in plain words
Missing, too short, too long, duplicate, invalid – each finding names the field, the measured value, the limit and what search engines make of it. Severities error, warning and hint; the findings list puts errors first.
Limits that match the result page
Meta title 60 characters, meta description 155: the defaults follow what search engines display, not what a shop will store. Every limit is a setting, and 0 switches its check off.
Duplicates per shop
Titles, meta texts and slugs are compared per marketplace and sub-account. Two listings behind one URL slug are an error, because only one of them is reachable under that path. On request the engine also reports similar meta descriptions.
Pictures judged like a buyer sees them
Minimum count and minimum edge length, measured on the shorter side – 2000 × 300 is a banner, not a large picture. Optionally one hint per listing when pictures carry no alt text – stating how many are described.
Score and history
Every audited listing starts at 100 points and loses 25 per error, 10 per warning and 3 per hint. A listing without findings gets its status line too – which is how “audited and clean” differs from “never looked at”. The curve on the dashboard shows the average of the full runs and today's state.
Focus keywords that are inherited
Rules check whether title, meta title and the start of the description carry the search term, and whether it is overused. Set it once per item category and every listing below inherits it; a listing's own keyword takes precedence. Channels in another language inherit only the keyword of their language. Without a keyword these rules stay silent.
Copilot writes. You decide.
The drafts are optional and run on your own company's Azure OpenAI resource. The model writes only from what your company knows about the listing – and no drafted listing text reaches a channel without your approval.
- Company data onlyItem description, attributes, brand, manufacturer, MPN, EAN and category, the current texts (the description up to 4,000 characters), the findings of the last audit and the focus keyword – with Search Console also the ten strongest search terms, and for duplicates up to five listings of the same shop with the same text. The prompt forbids invented data and instructs the model to write less rather than guess when the facts are thin.
- Six texts per listingTitle, meta title, meta description, keywords, slug and an HTML description – written against the same limits the audit checks them with afterwards.
- Current and proposed side by sideEvery draft becomes a content proposal, and every field can be edited before you apply it.
- Apply re-audits on the spotThe texts go to the hub fields, the listing is flagged for the next item update and audited again right away.
- Costs with a ceilingA run limit per batch, the monthly token budget of the AI setup, a stop after three failures in a row, and no second draft while one is still open.
What the AI does – and who decides
The audit works without AI. On top come exactly three Copilot features, each of which can be switched off under Copilot & AI Capabilities. All three run through the hub's central AI setup and count against its monthly token budget.
| Feature | What it produces | What a person does |
|---|---|---|
| SEO audit | No AI – pure rule evaluation inside your company | Reads the findings and works through them |
| 1 · Listing Content Draft | Six texts per listing – in the Copilot dialog, from the findings list, or as a batch for every listing with errors. Optional and off by default: a job that drafts for new listings in the background. For drafts in a foreign language, a literal translation for review on request, which is not saved | Apply or Keep it – otherwise no text reaches the listing. Automatic drafts wait as proposals too |
| 2 · Image Alt Texts (Preview) | A vision model looks at every gallery picture; a sentence is written only for pictures that demonstrably show their item. Needs a vision-capable deployment under Model per Feature – without one the feature does not fall back to the default model | Accept in the Alt Text Review. Only if an administrator turns on Accept Flawless Proposals Automatically (off by default) are flawless sentences written without that step |
| 3 · Marketing Text Translation | Translates the item card's marketing text into the languages of your sales channels – faithful or adapted to the market | Reviews and applies the translation; the item card's marketing text is never changed |
The proof comes from the search engine
Optional: connect your Search Console and the engine fetches impressions, clicks and search terms per page every night – through your own OAuth client, read-only. Or import a CSV export to get started.
Pages worth fixing
Impressions, clicks, click rate and position per listing, plus the missed clicks against your target click rate. That is what the list is sorted by – not by reach.
Focus keyword suggestions
Drawn from the terms people actually searched: the strongest term the page is within reach of, not the loudest one it ranks fortieth for. Accept writes it as the focus keyword.
What the new texts did
For every applied draft, the click rate before against after, in percentage points. The listings that got worse are in the same list.
Pages without a listing
Addresses Google reports that match no listing – with the reason. The engine calls the most-viewed ones itself and shows whether the shop redirects them or lets them run into a dead end.
What the SEO Engine does – and deliberately does not
The SEO Engine audits and writes in the hub; the connector transfers to the channel. That division of labour decides what the module leaves alone.
| Topic | Behaviour |
|---|---|
| SEO fields | Live channel-neutrally on the marketplace item in the hub; every connector maps them to its own channel fields (Shopware 6: metaTitle, metaDescription, keywords, seoPathInfo). Meta fields only reach the shop if the connector maps them |
| Channel profiles | What a channel does not take is not checked: an Amazon listing loses no points over a meta title it has nowhere to put |
| Languages | Listings on channels in another language are audited with the texts of their language; a missing translation is reported |
| Transfer to the channel | Not by the engine – applied texts travel with the connector's next item update |
| Listing texts without approval | Do not exist – there is no automatic mode that swaps listing texts in the background. For alt texts only if an administrator turns on auto-accept for flawless proposals (off by default) |
| Empty proposal fields | Delete nothing – an emptied field means “keep what is there” |
| Category texts and structured data (JSON-LD) | Not produced by the engine – both belong in the shop theme |
| Prices, stock, customer and vendor data | Never sent to the AI model |
One module, one price per tenant
The SEO Engine is available on Microsoft AppSource with a 30-day free trial.
€129 per 4 weeks
Net, per tenant, billed every 4 weeks from module activation. AI costs run through your own Azure OpenAI resource, the Search Console connection through your own Google project.
What you need
Business Central 28 or later (cloud, Essentials or Premium), the free hub app AD merchantCENTRAL and at least one marketplace or shop connector that carries listings.
Credentials kept safe
The Search Console client secret and refresh token live in the merchantCENTRAL credential store, never in a table, and cannot be read back once entered. The audit itself needs no credentials at all.
SEO in Business Central
Do I need AI to use the SEO Engine?
No. The audit is pure rule evaluation over the hub tables – no external service, no credentials, no tokens. Only the optional Copilot features need an Azure OpenAI resource, entered once in the central merchantCENTRAL AI setup; the alt texts also need a vision-capable deployment assigned there under Model per Feature.
Can Copilot publish texts without my approval?
Listing texts, never. Every draft becomes a content proposal; only Apply on the proposal card or Keep it in the Copilot dialog writes the texts to the listing. Automatic drafts end up as proposals too. Alt texts wait in the Alt Text Review for Accept – unless an administrator turns on Accept Flawless Proposals Automatically; it is off by default.
Does Copilot make up technical specifications?
The model receives your company's data only: item description, attributes, brand, manufacturer, MPN, EAN, category, the current texts, the findings of the last audit and the focus keyword. The prompt explicitly forbids anything else – no invented specification, figure, compatibility or certification – and allows technical data only from the item attributes. When the facts are thin, the model is instructed to write less rather than guess. Every draft is read by a person before it is applied.
What data leaves Business Central?
None during the audit. Your own company's Azure OpenAI resource receives item and listing data for drafts, the gallery pictures for alt texts (scaled down first, to 1,024 pixels by default) and, for translations, the marketing text together with the item description and manufacturer – never prices, stock, customer or vendor data. When figures are fetched, Google receives only the property URL and the date range; the page check calls addresses of your own shop. ALTENBRAND receives telemetry only: counts and durations per run, the decision on each proposal and error codes – no texts and no item numbers.
Which channels does the SEO Engine work with?
Every channel a merchantCENTRAL connector connects. The SEO fields live channel-neutrally in the hub, and the engine knows no connector. What a channel accepts at all is described in the marketplace's content profile, supplied by the connector – and that decides which rules apply to it. The meta fields only reach the shop if the connector maps them.
Do I need Google Search Console?
No, it is optional. Without it the search figures stay empty and the rules that depend on them stay silent. To get started you can import the per-page figures from a CSV export; the search terms per page come only through the API connection with your own OAuth client.
What does the SEO Engine cost, and what do I need to start?
€129 per 4 weeks, net, per tenant, with a 30-day free trial. You also need Business Central 28 or later (cloud, Essentials or Premium), the free hub app AD merchantCENTRAL, a connector with listings and a licence or running demo for the module SEOENGINE. The demo audits without limits; its AI drafts cover 20 products.
Manual, data sheet & online help
The user manual as a PDF – from setup through the rule set, Copilot drafts and Search Console to troubleshooting. Plus the online help on the individual pages of the module.
User manual (PDF, German)
Installation, licence, basic setup, rule set, focus keyword, daily work, score, Copilot drafts, Google Search Console, costs and data protection, troubleshooting.
Data sheet (PDF, German)
The short version on three pages: rule set, core features, system requirements and setup – to hand on to your department or management.
Online help
Help pages on initial setup, dashboard, findings, content proposals, Copilot drafts, alt texts and marketing text translation – in English and German.
See where your catalogue gives away search traffic
Install the hub and the SEO Engine from AppSource, check the rule limits and run the first audit across your catalogue. Free for 30 days.