§ The client
IJARST is a peer-reviewed academic journal publishing research across science and technology, running on a custom Laravel website. Like most journals, its entire reason to exist online is discoverability - researchers need to find its papers through Google, Google Scholar, and, increasingly, AI assistants that summarize the literature. The publishing was solid. The machine-readability was quietly broken.
§ Before / after: what machines saw
The agent's baseline audit scored the site's technical health 0/100 with six P0 (critical) blockers. The site wasn't badly designed - schema was present, robots.txt was clean, the paper pages were real. The damage came from a handful of sitewide defects that made the whole catalog unreadable to machines:
✕ Before
- ✕ Every page: identical title & description - 100 papers looked like 1 page
- ✕ 0 Google Scholar citation tags - papers unindexable as scholarship
- ✕ 6 critical (P0) blockers sitewide
- ✕ www + non-www duplicates, no canonical
- ✕ 15 H1s on the homepage · 14 images missing alt text
- ✕ Server response ~1.3s
- ✕ No llms.txt · entity health 24
✓ After - ~48 hours later
- ✓ Unique title & description on every page
- ✓ Scholar citation tags live - indexable scholarship
- ✓ 0 critical blockers
- ✓ Canonicalized - one authoritative version
- ✓ Clean structure - one H1, full alt text
- ✓ ~0.65s - 2× faster
- ✓ llms.txt published · entity health 35
§ Every fix moved the needle
This is the part that separates an autonomous agent from a dashboard: it didn't just report the problems - it fixed the code, wrote and edited the content, published it, re-crawled, and re-scored after each batch - with every change tested by both FirePencil's team and IJARST's own team before going live. The score climbed 0 → 82 → 92 → 98 as the blockers cleared - which is how we know each fix actually moved the needle rather than just looking done.
§ What the agent did - both tracks, owner-approved
SEO Agent track
owner-approved ✓
owner-approved ✓
owner-approved ✓
owner-approved ✓
owner-approved ✓
owner-approved ✓
AEO Agent track
owner-approved ✓
owner-approved ✓
owner-approved ✓
owner-approved ✓
owner-approved ✓
verified ✓
§ How every change shipped - the pipeline
No change - code or content - touched the live site without passing the same enterprise-grade pipeline. This is what “autonomous, but owner-approved” means in practice:
Every action across both tracks above carried this pipeline - which is why the “owner-approved” chips aren't decoration: they're the release gate.
§ The results: search visibility (30 days, Google Search Console)
30-day totals measured (Google Search Console). Weekly pacing illustrative - cumulative curves drawn to the measured totals.
§ The results: AI visibility (30-day multi-engine sweep)
Brand pickup observed in FirePencil's 30-day multi-engine sweep following the foundation fixes, with the strongest early AI-citability on Gemini.
§ The outcomes a journal actually cares about
Audit scores and impressions are the instrument panel - a journal runs on papers and authors. In June, the first full month on the repaired foundation:
“We publish solid research, but we had no idea that every page on our site looked identical to machines - or that Google Scholar couldn’t read our papers at all. FirePencil’s agent found it on day one and fixed it within two days, and our team reviewed and approved every change before it went live. The submissions and registrations we’re seeing now speak for themselves.”
§ Two clocks: site health vs AI visibility
These are different metrics on different clocks - and reading them together is the honest way to read this case study. Site health is the technical foundation: it moves in days once an agent is pointed at it. Overall AI visibility - how often engines actually surface and cite the brand - compounds slowly as indexation, entity signals and third-party evidence accumulate. Thirty days in, here is where both stand:
That ~14 is the number most vendors wouldn’t publish. We do, because it’s what a real visibility curve looks like at day 30 - the foundation work is what makes the climb possible, and the climb itself is a multi-month process. Anyone promising a 90/100 visibility score in a month is describing a different metric or a different reality.
§ The honest part: a foundation, not a finish line
Two days fixed the technical foundation - the fast, high-impact layer where a broken template or missing tags can crater an entire site. Getting cited as the authoritative answer across AI engines is the slower work that comes next: earning third-party mentions, deepening the entity, and answering more of the exact questions researchers ask. That's a multi-month effort, and it's exactly the ongoing, owner-approved work an autonomous agent is built to run. What this case study shows is how quickly the foundation can be repaired once an agent is pointed at it - and how much was silently lost while it was broken.
See your own before/after - free
Run FirePencil's free AEO audit and get your own baseline in about a minute: the real questions AI is asked about you, who it recommends instead, and the exact blockers holding your site back - the same first step that started IJARST at 0 and ended at 98.
§ Frequently asked questions
How long does AEO take to show results?
Three clocks, honestly: technical foundations can be repaired in days - IJARST went from a 0/100 to a 94–98/100 site health score in about 48 hours. Measurable AI visibility compounds over weeks - here, roughly 0 to 14/100 in the first 30 days across two optimization sprints. Full citation authority is a multi-month effort. Anyone promising high AI-visibility scores within a month is describing a different metric or a different reality.
Why was IJARST not showing in Google Scholar?
Its pages carried none of the Highwire/Google Scholar citation meta tags (citation_title, citation_author, citation_date and related fields) that Scholar requires to read a paper’s title, authors, journal and publication data. Without those tags, Scholar cannot reliably index articles as scholarship. FirePencil’s agent added the citation tags across the paper pages, making the journal indexable - the single highest-impact fix available to any academic publisher.
What does an autonomous AEO agent actually do?
It executes rather than reports: it audits the site against answer-engine readiness checks, fixes the code (templates, canonical tags, structure, performance), writes and edits content, publishes through a staged pipeline, then re-crawls and re-scores to verify each fix worked. In this case that meant three fix batches in 48 hours - score climbing 0 → 82 → 92 → 98 - followed by ongoing optimization sprints. Every action is approved by the site owner before it ships.
Is it safe to let an AI agent change a live website?
Safer than most manual processes, by design: the site is fully backed up before any batch; every change is built in staging, never on the live site; each sprint is tested by both FirePencil’s QA and the client’s own team; nothing deploys without one-click owner approval (human-in-the-loop); and the pre-batch backup makes rollback available at any moment. IJARST’s team reviewed and approved every change in this case study before it went live.
Can these AEO results be verified?
Yes. Search metrics can be cross-checked publicly through third-party dashboards such as Semrush or SE Ranking - at the time of writing, Semrush’s independent AI Visibility reading for ijarst.com matched the ~14 figure reported here. The audit logs, fix batches, and case-study KPIs are verifiable on request via [email protected].
Case study data is drawn from FirePencil's own audit and monitoring records for IJARST (May–July 2026). Site health scores reflect FirePencil's AI-first readiness model (the 12 AEO pillars) - deliberately different from legacy-SEO health checklists used by tools like Semrush, so absolute scores differ by design; outcome metrics (visibility, citations) are the layer where third-party dashboards and our reporting can be directly compared - and agree; the Overall AI Visibility figure (~14/100) is approximate, from FirePencil's multi-engine sweep; search metrics are from Google Search Console; weekly chart pacing is illustrative with cumulative curves drawn to measured 30-day totals. Business outcomes (submissions, registrations) are client-reported figures for June 2026, the first full month after the fixes; multiple factors contribute to such outcomes. Results describe one client's experience and are not a guarantee of any specific outcome. Third-party names (Google, Google Scholar, ChatGPT, Gemini, Perplexity, Claude) are trademarks of their respective owners; use is descriptive.