The honest version, up front: we run our GEO program on our own site, we measure it with live AI probes, and we're going to show you what actually happened — including what hasn't worked yet. Too many "GEO case studies" are victory laps with no baseline and no method. This one has both: a measured starting point, the specific things we implemented, what moved and what didn't, and the honest timeline (spoiler: the infrastructure is fast, the authority is slow). If you want to know what running GEO on a real site actually looks like, this is it.
The baseline: measuring before optimizing
We started where every honest GEO effort should — by measuring. We probed the AI engines with our real buyer questions ("best AI automation agency for a business that wants to own its system," and similar) and logged the answers. The baseline was humbling and instructive: for our core money question, we were not named in ChatGPT or Perplexity, and when Perplexity answered, every source it cited was a roundup listicle — mostly published by agencies listing themselves. That single measurement was worth more than a month of guessing: it told us exactly why we were invisible (not in the cited sources) and exactly what to build (citable assets and presence in those roundups).
What we implemented
With the baseline in hand, we worked the layers of a real GEO program, in order:
- Entity & schema: coherent Organization, Service, and FAQ structured data, with consistent identity across our profiles, so machines resolve who we are with confidence.
- Machine-readable positioning: a maintained llms.txt carrying our positioning, facts, and key guides in the format AI assistants ingest — auto-updated from our best content.
- Citable content: stat-first, dated, sourced pages — verified pricing, original comparison data, direct answers to the exact questions buyers ask the engines.
- Index plumbing: AI-crawler access, IndexNow submission into Bing (which ChatGPT search reads), a clean sitemap, and canonical hygiene.
- Off-site strategy: a plan to earn placement in the roundup listicles the engines actually cite — the piece we found mattered most.
What moved, and what didn't
Here's the honest split. The infrastructure moved fast: schema validated, llms.txt served and fetched, pages indexed in Bing within days, IndexNow pinging on publish. Those are inputs we control, and they were in place quickly. The retrieval-layer signals — being pulled into a generated answer — are the medium-term game, showing early movement as citable content matures and gets indexed. The authority layer — actually being named by the engines for our competitive money question — is the slow part, because it depends on off-site corroboration (reviews, mentions, placement in cited sources) and on our domain's trust recovering from a prior thin-content issue. That doesn't happen in weeks; it compounds over quarters.
The most valuable lesson: GEO isn't one lever, it's a stack with different time constants. The on-site work (schema, llms.txt, citable content, indexing) is fast and fully in your control — do it and it's done. The off-site work (being cited and corroborated across the web) is slow and only partly in your control — it's earned, not configured. Anyone promising fast AI-answer dominance is either ignoring the authority layer or lying about it. The realistic picture is: quick infrastructure wins, then patient authority-building.
The honest timeline
| Layer | What it is | Time to move |
|---|---|---|
| Infrastructure | Schema, llms.txt, indexing, IndexNow | Days to weeks (you control it) |
| Retrieval content | Being pulled into generated answers | Weeks to a couple months |
| Authority / citation | Being named for competitive queries | One to two quarters+ (earned) |
What we'd tell anyone starting GEO
Measure first — you can't manage an AI answer you haven't read. Do the on-site infrastructure immediately, because it's fast, cheap, and fully in your control. Then commit to the slow, unglamorous authority work — reviews, citable content, earning mentions in the sources engines trust — because that's what actually gets you named, and it's the part everyone wants to skip. And keep probing monthly, so progress is a logged fact rather than a hope. That's the method, run transparently on our own site, and it's exactly what we run for clients.
What we're doing next (and why we're sharing it)
We're being public about this because the honest version is more useful than a polished one — and because the next phase is the hard part everyone skips. Having done the on-site infrastructure, our focus now is the authority layer: earning reviews (which feed the directories AI cites), getting into the roundup listicles our probes showed the engines pulling from, and letting our domain's trust continue to recover. These are slow, unglamorous, partly-out-of-our-control levers, and that's precisely why they're where the real competitive advantage is — most competitors won't do the patient work. We'll keep probing monthly and updating what moves.
The broader point for anyone reading: a GEO case study without a baseline, a method, and an honest account of the slow parts isn't a case study, it's an ad. What makes this one worth your time is that it shows the real shape of the work — fast infrastructure wins that feel great, followed by the patient authority-building that actually determines whether an AI names you for a competitive query. If a provider promises you'll dominate AI answers in weeks, this case study is your evidence to be skeptical. The infrastructure is weeks; the authority is quarters. Both are worth doing; only one is fast.
Did you actually get cited by AI?
We're transparent that it's a stack with different speeds: the on-site infrastructure was in place within days to weeks, retrieval-layer signals are showing early movement as citable content matures, and being reliably named for our most competitive money query is the slow, in-progress authority work measured over quarters. We share the real, ongoing picture rather than claiming a finished victory — because the honest timeline is the useful lesson.
Why share your GEO strategy publicly?
Two reasons. First, transparency is our brand — showing the real method (including the slow parts) builds more trust than a polished victory lap. Second, the competitive moat in GEO isn't secret knowledge; it's the patient execution of authority-building that most won't do. Sharing the method costs us little and demonstrates the expertise, which is itself good GEO. The concepts are public anyway; the execution is the differentiator.
What would you do differently starting over?
Very little on sequence — measure first, do the fast on-site infrastructure immediately, then commit to the slow authority work — but we'd start the off-site authority levers (reviews, roundup placement) even earlier, because they have the longest time constant. The lesson is that the slowest-moving lever should be started first, since it takes the longest to pay off. Front-load the patient work; the fast work stays fast whenever you do it.
FAQ
How long does GEO take to work?
It's a stack with different speeds. On-site infrastructure (schema, llms.txt, indexing) moves in days to weeks because you control it. Being pulled into generated answers follows over weeks to a couple months as citable content matures. Being reliably named for competitive queries takes one to two quarters or more, because it depends on off-site corroboration and domain trust — the earned, slow part. Fast infrastructure, patient authority.
What's the single most important GEO step?
Measuring first. Probing the AI engines with your real buyer questions and logging whether you're named, cited, and who wins instead tells you exactly where you stand and what to build. Our own baseline (not named; competitors cited via roundup listicles) directly shaped our entire strategy. You can't optimize an answer you haven't read — measurement turns GEO from guessing into a task list.
Can any business get cited by AI?
Most can, with the right work and realistic patience — but the timeline depends on your starting authority and competition. A business in a low-competition niche with clean infrastructure can appear relatively quickly; one competing for a contested money term against high-authority sources needs the slower off-site and trust-building work. The on-site part is winnable by anyone; the competitive citation part is earned over time.
Is this something I can do myself?
The on-site infrastructure — schema, llms.txt, indexing, citable content — is doable yourself if you're technical and patient, and the measurement method is free to run. What a service adds is execution at quality and the off-site authority work that's hardest to do alone. Start with the free measurement and the on-site fundamentals; bring in help for the slow, competitive authority layer if you want to accelerate it.
How do you measure GEO progress objectively?
With a frozen set of real buyer questions, probed across the AI engines monthly, logging for each whether we're named, whether we're cited, and which sources win instead. Comparing month over month turns 'are we winning at AI?' from a feeling into a tracked number and a task list. We also watch supporting signals — Bing indexation, AI-Overview citations, and referral traffic from AI surfaces — but the probe log is the objective backbone.
What's the biggest GEO mistake you see businesses make?
Skipping measurement and chasing the fast, visible infrastructure while ignoring the slow authority work that actually gets you named. Plenty of businesses add schema and an llms.txt, see nothing happen in a week, and conclude GEO doesn't work — when the reality is the authority layer takes quarters and they never started it. The fix is to measure from day one and commit early to the patient off-site work, not just the quick on-site wins.
Want this run on your site — measured, transparent, and owned by you? Our GEO service starts with the same probe-based audit we ran on ourselves. Related: how to measure AI visibility and what GEO is.