One brand, three different SEO strategies — depending on which provider you're trying to win.
Empirical finding from running this across multiple sites: the three major AI surfaces ground their answers on three meaningfully different source types. What wins on one barely shows up on the others. If you only measure one, you're playing a third of the actual game.
OpenAI
Web-grounded responses via the Responses API + web_search_preview tool.
Gemini
Google's Gemini models with Google Search grounding enabled.
Google AI Overview
The AI summary at the top of Google search results. Captured via SerpApi.
Three position metrics, captured per provider, every scan.
Knowing whether you "appear" isn't enough. The mention has to be in the right shape — early enough in the answer to be read, recommended explicitly enough to convert, cited authoritatively enough to drive click-through. We measure all three.
Text position
Where your brand appears in the model's free-form answer text. Earlier = more likely to be read; sometimes the only mention is the one in the first sentence.
Recommendation rank
If the model produces a recommendation list (top picks, finalists, suggestions), where do you sit? Position 1 is dramatically more valuable than position 5.
Citation rank
In the model's cited sources, where does your domain appear? First citation gets ~70% of the click-through. Citation #4 effectively gets none.
Each scan also captures the full response text and the complete list of cited sources, so you can see exactly what the model said about you and what it grounded on. No more guessing.
Define the queries. Schedule the scans. Get the reports.
1. Query design
We work with you to define a high-priority query set — the questions a real customer would ask an AI about your business. Typically 10–25 queries: identity-level ("who is X"), comparative ("X vs Y"), intent-driven ("best X in Z"), and edge cases that might attract fabrication.
2. Scheduled scans
The detector runs weekly — by default Monday morning, your timezone. Each scan runs every query through all three providers, captures responses and cited sources, computes the three position metrics, and writes the deltas (new mentions, lost mentions, position changes) to a per-site database.
3. Weekly delivered report
Friday morning you get a markdown report by email. Top of the report: anything that changed this week, with links to the actual provider responses. Below that: a summary of where you stand across all three providers on all your priority queries. Trend chart over the last 12 weeks. Recommended next interventions.
4. Recommended interventions
For each issue identified — fabrication risk, missed citation, dropped rank — the report recommends a specific content intervention. Most are short (an article, a page update, a Q&A). Some are structural (schema.org work, sameAs additions). All are scoped concretely so you can decide whether to do them yourself, hand them to us, or ignore.
Indexing latency for new long-form content is 48–72 hours. Publish the recommended fix on Monday, see the model citing it on Wednesday or Thursday — verified by the next scan, with deltas visible in the following Friday's report.
Per site. Per month. No setup fee.
Cancel any time. Historical scan data stays available for 90 days post-cancellation in case you want to re-attach later. No setup fee on any tier.
Either you've got a site we built, or you don't. Both work.
The monitoring is a standalone service — it watches whatever site you point it at. You don't need to be a rebuild client to use it. Around half of monitoring clients are running on WordPress, Shopify, Squarespace, or hand-rolled stacks; the monitoring is platform-agnostic.
Where it gets more powerful is when we built the site too. Site we built + monitoring = closed loop: we see the issue surface, we make the content change, we watch the model pick it up within 48–72 hours, we report the lift. That's the loop the Miabella case study documents.
Every client gets a private repo. And an agent on top of it.
Reports are a snapshot. You skim the top, close the tab, and by Tuesday the question you actually have isn't covered. Your GA dashboard shows an uplift three months later and nobody can say why — the trail of page changes, bot crawls, and citation deltas is scattered or gone.
For every monitoring client, three streams write continuously into a private git repo: every site change you ship (with diffs), every AI crawler visit (GPTBot, Google-Extended, PerplexityBot, Bingbot — timestamped), and every weekly scan delta. Same repo, same timeline. The "why" question stops being archaeology.
Sitting on top of the repo: an agent you talk to in plain English. "What did we change in the last month that moved anything on Gemini?" — and you get the specific changes, the specific queries that moved, the specific evidence. Not a summary. The thing itself.
Test it out — the chat on this page knows everything about this site. Ask about pricing, about how the monitoring works, about which Cloudflare features we disabled and why. And it probably talks your language — ask in Afrikaans, French, Mandarin, Welsh, whatever, and it'll answer in kind. That's the same shape every client gets, scoped to their own data.
That's the short version. The full argument — why a snapshot report was always a thin substitute, what changes when the storage layer is a git repo with an agent on top, and the causal chain you can actually see from page edit to crawler hit to citation move — is the read at Reports tell you what changed. They don't tell you why →
Real results, in plain English. Traceable from change → crawl → citation → metric. Included on every monitoring tier.