Why your site ranks on Google but AI never mentions it

BrandBrahma's AI agent, Perceive, Plan, Act, Verify, fixed 31 to 68 AEO score on Meridian Ledger's site, with 14 of 19 image alt texts approved and 6 fix types written, while 6 proposed edits were rejected, including a deliberate canonical URL. The agent wrote directly into WordPress via Rank Math fields and re-scanned pages to measure improvements.

From reports to fixes that shipped Meridian Ledger again — the bookkeeping SaaS from the first story, now past the diagnosis and into the part where something has to change on the site. Three audits had told them what was wrong. That is where most tools stop, and it is also where most fix lists quietly die — the founder was the only person who could edit the site, and the list was thirty items long. Perceive. The agent scanned the live URL across all four lenses and returned a per-signal breakdown, not a grade. Plan. Findings became a ranked list of concrete edits, each showing the exact current value and the exact proposed value — judgeable without reading any code. Act. Each edit sat behind an approval gate. Approved ones were written straight into the WordPress post, through Rank Math's own fields rather than around them, so the values appeared in the interface they already used. Verify. It re-scanned the same page and diffed it against the original snapshot — so the improvement was measured rather than asserted. ✓ Meta title 58 chars, intent-led, written to Rank Math's title field APPROVED ✓ Meta description Rewritten to survive truncation in the results page APPROVED ✓ H1 heading Two competing H1s reduced to one clear page topic APPROVED ✓ JSON-LD schema FAQPage + Organization injected — the AEO gap from story one APPROVED Image alt text 14 of 19 approved; 5 rejected as inaccurate and edited by hand PARTIAL ✕ Canonical URL Rejected — deliberate cross-domain canonical the agent misread REJECTED They rejected things, and that was fine. Six of the proposed edits did not survive review. The canonical in particular was a deliberate decision the agent had no way of knowing about. An approval gate is only meaningful if rejecting is a normal outcome. Every edit recorded its previous value. Nothing written was unrecoverable, which is what made approving a batch of fourteen alt-text changes a five-minute job instead of a careful afternoon. 4 Credits, full cycle 31 → 68 AEO after verify, illustrative 6 Fix types written The transferable lesson The gap between knowing what is wrong and having it fixed is where most SEO work dies. Judge any agent on two things: whether you can reject individual changes, and whether it re-measures afterwards. Without the first you are not in control; without the second nobody can tell you whether it worked. Meet the agent https://www.brandbrahma.com/ai-search-visibility/