What’s new at Specoria
A dated list of what we shipped. Items are live unless marked “Built · not open yet”: those exist in our code but aren’t open to customers, and each says what it waits for.
Items are written by hand from our commit history. The date is the day of the commit; the short codes are the commits behind each item.
A multi-client panel for agencies
Teams with an agency account see their linked client projects in one table: score, level, 30-day change, critical blockers, last re-check, plan and alerts. You add a new client by domain; the client only gets an email if you invite them. The client can remove the agency's access from their own panel at any time. The executive report opens with the agency's name and logo. Our team opens agency accounts.
See it → · Commits: b432037
Agent Showcase: a UCP profile, store facts and llms.txt from your verified data
On paid plans, the panel's new Agent Showcase tab builds three files only from the scan, the catalog and values you approved: a UCP profile, a store facts JSON with source and date, and an llms.txt. You review and publish them; unpublishing makes the addresses return 404, and you can go back to an earlier version. No checkout capability is declared and no field we couldn't verify is written. A daily health check confirms a breakage with a second check and lets you know.
See it → · Commits: 8084e35
Agent traffic: IP verification and continuous collection
A user agent can be faked; a provider's published IP ranges can't. The agent traffic report now checks a bot's claim against the published IP ranges of its provider (OpenAI, Anthropic, Perplexity, Google, Apple and others); providers without a list stay "can't verify". On paid plans you can send logs continuously instead of uploading files, with Cloudflare Logpush or a small script; full IPs aren't stored, and the report is compiled daily.
See it → · Commits: fdd4bff
A readiness score per product and an OpenAI product feed
The panel catalog gives each product a readiness score on six axes: price, buyability, identity (GTIN, brand), stock, attributes and returns; an axis that can't be measured leaves the denominator. An OpenAI product feed is generated from your catalog and validated against the specification; you download it or share its stable URL. We never send the feed anywhere ourselves, and Türkiye isn't in OpenAI's program yet.
See it → · Commits: adf9037 7f072af
ChatGPT Shopping add-on: what it measures and what it can't
A new page explains plainly what the ChatGPT Shopping add-on does: we ask an OpenAI model, through the API with web search, the shopping questions your customers ask, and record whether it recommends you, who gets the purchase link and whether it states your price correctly. Every number is labelled "API measurement".
- What we can't measure is on the same page: API answers have no real product cards, your visibility in the app may differ, and we never scrape the ChatGPT app.
- An OpenAI product feed (JSONL and a Google-compatible TSV) generated from your catalog and validated row by row is included as readiness in every plan.
- There is no ChatGPT product feed program for Türkiye yet; the page says so too.
See it → · Commits: d5adff9 b2ba7e0
Tool comparison, hotel and service pages, trust center, statistics
- Specoria and other tools: an honest comparison of readiness tests, visibility tools, content enrichment and agencies; no competitor names, including the rows where Specoria says "no" or "partly", with a dated source line.
- For hotels and service businesses: what the free test and the simulator do on those site types, with their checks and weights.
- Trust center: data handling, sub-processors, AI use principles and a security contact.
- Agentic commerce statistics: every figure with its source, date and scope, plus a ready-to-cite box.
See it → · Commits: 40965b7 172cd92
A screenshot-only agent in the full report
The agent simulator's full report has a new lane: a computer-use agent that sees the page only as a screenshot. ChatGPT agent and Claude's and Gemini's computer-use agents perceive pages this way, so this lane can get stuck in different places from agents that read the page structure.
- The moment the agent got stuck is replayed from saved screenshots.
- The safety rules are the same in every lane: agents stop at checkout, never order and never type personal data.
See it → · Commits: 9d6f7b5 252e69b
Content Studio (built, not on sale yet)
Content Studio drafts copy for products with missing attributes or weak descriptions, only from the store's own data and with numbered source lines; any number, claim or value that isn't in the sources is removed and marked "input needed". You edit, approve and export the drafts. Not open yet: until its price is approved, only our team can see it.
Commits: 8c8fd49
Our crawler identifies itself as SpecoriaBot
In the free test, the request that identifies itself as a bot now uses only the SpecoriaBot identity; we don't imitate any other company's bot. The crawler policy page lists the requests we make, the time budget, the cache and how to have your site excluded; the figures come straight from the code.
See it → · Commits: a72b859
Free test v2.4: protocol discovery, policy links, rate limits handled fairly
With score model v2.4 the free test reads more; most of the new lines are informational and don't affect the score:
- Agent protocols and discovery files: the UCP profile (and common setup mistakes), the MCP discovery file, the A2A agent card, AI Catalog, API Catalog, the Web Bot Auth key directory, WebMCP markers in the page and a possible ACP checkout endpoint. Only GET requests are used; no endpoint is called and no session is opened.
- Policy and contact links: the eight link types ACP asks merchants for.
- Shipping and returns detail: delivery times, shipping regions and the return type.
- Rate limits aren't counted as blocks: if a site returns 429, we follow Retry-After and retry once; a check we couldn't measure shows as "couldn't verify", never as a fail.
See it → · Commits: 10918f8 a1047a8 172cd92
GA4 and Search Console connection (built, not open yet)
A read-only connection to Google Analytics 4 and Search Console: AI-referred sessions, key events and revenue, and Search Console's last 28 days against the previous 28, shown on the agent traffic page. Removing the connection also revokes the permission at Google. Not open yet: it becomes available in the panel once we switch on the Google connection.
Commits: 82b820a
Glossary and formulas, platform fix guides, free tools
- Glossary and formulas: score, partial credit, critical blocker, levels, potential score, crawl-to-refer, mention rate and share of voice, defined the way we compute them; the worked example comes from the real sample report.
- Platform fix guides: where to make each fix on Shopify, ikas, Ticimax, IdeaSoft, T-Soft and WooCommerce. They are read-only guides; we don't connect to your store's admin.
- Free tools: a robots.txt tester for AI bots, an AI bot user-agent identifier and a crawl-to-refer calculator for a server log. No sign-up; nothing you paste is sent to us, it all runs in your browser.
See it → · Commits: ed26dcd
The free test recognises hotel and service sites
The free test recognises the type of site from its home page (without extra requests). On hotel and service sites, checks that only make sense for a store (cart, guest checkout, product feed and so on) are marked "not applicable" and leave the score; the type's own checks are measured instead. The result page and the method page speak to the site type too.
See it → · Commits: 23db5a0 6d9a90f 334c1e5
Agent Readiness Index 2027: free web summary and a PDF with no form
The index results can now be read without a form: the level distribution, the "fully ready" funnel, tables by sector, platform and check, the method and the limits of the sample. Every figure comes from the published PDF, with its page number; the aggregated data downloads as CSV under Creative Commons Attribution 4.0, with no store names. For now the web summary carries the figures of the English-language edition (471 stores); the Türkiye edition's results are in the Turkish PDF. The PDF report now downloads without a form too; anyone who wants to hear about the next edition can leave just an email.
See it → · Commits: e375dcd
Marketplace listing readiness: Trendyol (built, not open yet)
A read-only connection with your own Trendyol seller API key: your listings are matched to your catalog by barcode or GTIN first, then by stock code; price and stock conflicts, GTIN problems, title quality and empty required attributes become tasks. Marketplace pages are never scraped and your listings are never changed. Not open yet: it opens in the panel once our legal review is complete. Hepsiburada depends on integrator registration and is on the roadmap.
Commits: 752e310
A page for every check and a public sample report
- Every check in the free test has its own permanent page: why it matters, how to fix it, example code, its weight, whether it is a critical blocker and which site types it applies to. Weights are read from the score model, not typed by hand.
- The sample report shows the real, dated free-test result of a public mock store (web-scraping.dev), unedited; the same result can be downloaded as JSON.
- The method page gained a coverage table showing what an automatic scan can't see.
See it → · Commits: ce90ffb 5512cf5
A question library for the visibility check
You can now pick the questions for the AI visibility check from a ready-made question library: shopping question templates in Turkish and English for eight sectors, filled in with your store's product types. The questions you pick are asked first in the measurement; the limit of 10 custom questions per project stays the same.
See it → · Commits: 8c8fd49
Scan API, an MCP server for your team and a GitHub Action
You can run the free test from your own scripts or CI: POST https://app.specoria.com/v1/scan runs the same test and returns the result as JSON. The MCP server at POST /v1/mcp gives an AI assistant three tools: scan a store, list your projects and read one project's tasks. You create the key in your panel; we store only its SHA-256 hash, and every key has hourly and daily limits. The GitHub Action fails the step when the score is below your threshold; it isn't listed on the GitHub Marketplace yet, and we send the files on request.
See it → · Commits: e18c7ce
Sentinel and threshold alerts
On paid plans the panel runs a light daily check without AI and tells you when something breaks: the score drops by 5 points or more, a passing check fails, robots.txt starts blocking a new AI bot, product structured data breaks, the home page hits bot protection, or a simulation no longer reaches its goal. An alert is confirmed by a second check and shows the dates of both measurements. It goes to email, Slack and Microsoft Teams.
See it → · Commits: a9614ff eb90bb7
Share your free test result, and remove the link when you want
Your free test result no longer has to live only in your browser. If you want, the result page creates a share link: a 30-day copy of the result opens, the page is closed to search engines, and the shared page's title shows the domain, score and level.
- Whoever creates the link can remove it before it expires; a removed link no longer shows the result.
- Sharing is optional; if you don't create a link, the result isn't tied to you on our server.
See it → · Commits: 6ce7d4d 89c43e1
Journey metrics, step playback and the Turkish checkout journey in the simulator report
The agent simulator's report goes deeper:
- Journey metrics: number of steps, time, pages visited, back-tracks and how the agent reached the product (search, menu or direct), plus a step tree with its branches.
- Step-by-step playback from saved screenshots.
- The WebMCP tools the page offers.
- The Turkish checkout journey: when an agent reaches checkout, items such as forced sign-up, province and district selection, a Turkish ID number field, SMS verification, 3D Secure, cash on delivery and bank transfer are only observed; nothing is typed into any field.
See it → · Commits: 1fdb8d9 125791b
Specoria Commerce Readiness Specification 1.0
Every scored check of the free test is now in one specification: its stable ID (SC-AREA-NN), level (critical blocker or recommended), weight, requirement and how it is tested. It is generated from the scoring model itself and also published as machine-readable JSON; a mapping to AgentReady items is on its own page.
See it → · Commits: e2523b5
Writing fixes back to your store: Shopify and WooCommerce (built, not open yet)
Optional write-back applies suggested fixes, such as structured data, to your store one at a time: each change only with your approval, with a diff to review and one-click undo to the previous value. The value in the store is re-read right before writing, and the score changes only through re-measurement. Not open yet: it becomes available once we switch on the store apps. Until then, you apply the ready-made code yourself.
Commits: 245af75
Test form: marketplace addresses redirected, measured timing shown
- If you enter a marketplace listing address, the test doesn't run and your free scan isn't used; you're asked for your store's own site. A marketplace listing check is on the roadmap; we don't write as if it exists today.
- Under the form: "No sign-up · no email · no card", plus the median scan time of the last 30 days when there are enough measurements.
See it → · Commits: 6ce7d4d
Repeated sampling and an accuracy check in AI visibility
AI answers vary from run to run. On paid plans each question is asked 3 times per engine in a measurement, and the report says "mentioned in 2 of 3 runs" with a consistency rate. The monthly cap is never exceeded: if all runs don't fit, the number of runs is lowered and the report says so. Price, stock, shipping and returns in the answers are compared with your store's data; a difference seen in at least two answers becomes a task.
See it → · Commits: ee05ec5 61c7883