Run a GEO Report
See how clearly AI answer engines and shopping agents can read your storefront, understand the scores and confidence, and fix the highest-value gaps first.
The GEO Report (Generative Engine Optimization) shows how well AI answer engines — ChatGPT search, Perplexity, Claude, Google's AI features — and AI shopping agents can reach, understand, and trust the pages on your storefront. It replaces the older Audit Report and Opportunities page with plain-language categories, a 0–100 score for every check, and a separate confidence value so you know how sure Forge is.
The report measures readiness. It does not predict or promise citations or rankings; the Citations module tracks what AI answers actually say.
Open the report
- Open AEO & Agentic Storefronts and choose the storefront in the switcher (for example, your production site or your staging site).
- Select the GEO Report tab.
- If no report exists yet, select Run GEO Report.
Run a report
The Run GEO Report window has three steps.
If the storefront has no managed pages yet, choose Choose Scope & Run First Scan inside the window. Discovery shows the site's content types and URL groups; it can take a minute. Choose what to include, then select Save Scope & Run Scan. Wait for the inventory scan to finish. The window then reloads the discovered pages so you can continue. Forge will not start a report with zero eligible pages.
You can also start this scan from Pages → Choose Scope & Run First Scan. If discovery fails, check the connection and use Retry Scope Discovery.
Pages. Keep all managed pages or pick sections. Forge automatically skips pages that usually should not be scored — cart, checkout and account pages, legal and policy pages, search results, tag and date archives, attachments, and brand-asset or media-kit pages — and lists them with the reason. Tick any of them to include it anyway.
How to judge. Choose one option:
| Option | What it costs | What you get |
|---|---|---|
| Rules Only | Free | Every check Forge can measure automatically. Checks that need judgment show as Not measured. |
| Forge-funded AI (for example GPT-6 Luna or Grok 4.7) | Ingots — the window shows the expected amount and the maximum | Full report. Forge reserves the maximum before it starts and charges only what the run actually used. |
| Your own provider key | Billed to your provider account | Full report using the key you added in Settings › Connections. |
Review. Confirm the page count, the expected number of AI-judged pages, and the estimate, then select Run GEO Report. Progress appears on the report page and in the Tasks menu.
Why large catalogs stay inexpensive
Many product pages differ only by size or configuration. Forge groups near-identical pages, judges one representative (plus any page that differs in its specifications and a random check page), and shares that judgment across the group. Every page still gets its own automatic checks. Pages that have not changed since the last run reuse the earlier judgment at no cost.
Read the results
Overall score and grade. A (85–100) Strong, B (70–84) Good, C (50–69) Developing, D (25–49) Weak, F (0–24) Critical. Large groups of similar pages are balanced so hundreds of near-identical products do not drown out your capability, industry, and calculator pages.
Confidence. High, Moderate, or Low, shown next to every score. It reflects how much evidence Forge could read and how the result was judged — a page judged directly is more certain than one that inherited its group's judgment. A report marked Provisional could not measure enough checks yet; fix access problems or re-run.
Access blockers appear first. If AI search crawlers are blocked or pages cannot be indexed, fix that before anything else — no content improvement helps a page an engine cannot read.
Categories. Eight questions, each with its own score:
- Can AI Engines Reach It?
- Does It Answer Buyer Questions?
- Are Facts and Entities Clear?
- Can Buyers Evaluate the Product?
- Can Machines Interpret It?
- Can Readers Verify It?
- Can Useful Pages Be Found?
- Can an Agent Take the Next Step?
Select a category to filter the pages list.
Top Improvements ranks the fixes worth doing first by impact, effort, and confidence. Each one says why it matters, how to fix it, how many pages it affects, and links to those pages. Items Forge is unsure about appear under Investigate so you can check them yourself.
Pages. Sort and filter every scored page. Open a page to see each check with its score, why it matters, how to fix it, and the page text the judgment was based on. Pages marked Inherited from similar pages share their group's judgment. Use Export CSV to share the full list.
If the storefront scan fails
A scan failure appears on the report with a reason, such as rejected WordPress pairing, an HTTP error, or an unreachable site. Open Settings → Connections → Storefront Integration Settings, choose the affected site's connection, and use Re-Pair Connection with a fresh code from WordPress Settings → ForgeAEO → Generate Pairing Code. Re-pairing preserves reports, inventory, catalog, families and settings. Validate the repaired connection and start a new report.
Reset Integration permanently deletes that storefront and its reports, inventory, catalog, families and related drafts. Use re-pairing for a pairing repair; reset is for deliberately discarding the connection and its data.
If the SI analysis did not run
A Forge-funded or your-own-key report can finish Complete with gaps with a note that the SI (AI) analysis did not run — usually because the AI provider account had no credit or the key was rejected. The automatic checks are still complete and scored, so the access blockers and structural findings are real. Fix the provider account or key, then press Run SI Analysis Again on the report. Forge reuses the pages it already read and only repeats the judging step, so you are not charged for reading the site twice and the new report appears within a few minutes. The reason is also recorded in Settings › Activity Log.
Run it from your AI client (Forge MCP)
If you use Claude or another MCP-capable AI client, connect Forge MCP (Settings › Connections › MCP) and ask your assistant to run the GEO Report for a store, for example: "Run a GEO Report on our production storefront and summarize the top five fixes."
In this mode your AI client does the judging with its own model, so the run uses no Forge AI and no Ingots. Forge still collects the page evidence, runs every automatic check, validates what the assistant submits, and calculates the scores, so the report looks the same in the web app. Your assistant can then read the findings and create draft fixes — new copy, FAQ answers, specification tables, or JSON-LD — for your team to review. Drafts are never published without review.
Platform staff working across several customer accounts pick the account first, then the storefront.
Confirm the outcome
After you publish fixes, re-run the report for the affected pages or sections. Unchanged pages reuse earlier judgments, so a focused re-run is quick and inexpensive, and the category and page scores show what moved.