omniagent // geo · aeo
People increasingly get their answer without opening a single link. We fill your blog with material built for how ChatGPT, Perplexity, Claude and AI Overviews pick a source — and every piece is checked by code, not by eye.
Classic optimisation fights for a place in the results. Answer optimisation fights to be inside the answer someone reads instead of the results. Those are different demands on the text — they change its structure, not just its headings.
The engines pick sources differently, which makes «optimise for AI» a meaningless sentence. What can be optimised is each engine's actual mechanism.
Pulls from Bing's organic index and its own crawl. So the first move is not Google but Bing plus instant indexation — for a new domain that is the cheapest door there is.
Retrieval over primary sources; values recency most and reads llms.txt. So the piece has to be self-contained and dated.
Favours long analytical writing with a named author. So the article is signed by a person and stands on its own data rather than on a retelling.
Weigh YouTube and discussion heaviest. So an article needs a video — which is exactly what we already have on the line.
Eleven stages, each with its own artifact and its own condition to pass. Skipping one is not possible: the next simply has nothing to work from.
Candidates from two independent demand signals: what people type, and what they actually argue about.
gate: The two signals are never blended into one score
Selection by winnability: opportunity × urgency against the strength of the domains already holding the topic.
gate: A topic with unmeasured demand cannot enter the plan
Entities, the answer-first block, the category, the author, your own evidence and the CTA the piece earns.
gate: At least one piece of your own evidence + your approval
The heading plan: every section with a job, questions pulled into their own blocks.
—
The text: answer in the first paragraph, self-contained 40–60 word blocks, Q&A, citations.
gate: ≥3 external sources, ≥4 internal links, zero uncited statistics
A de-slop pass — machine tells out, meaning untouched.
—
The companion video, through the same pipeline that makes your reels.
gate: The hook quality gate + a cost preview before and the actual after
The page treatment: layout, art direction, the embedded video.
gate: Your brand tokens only
The article and the page as actually served.
gate: 14 checks + the speed budget — both halves block
Publishing with schema, an IndexNow ping, backlinks across siblings, the category hub refreshed.
gate: Every internal link has to resolve
Trimmed versions on external platforms linking back; the video to YouTube.
—
Almost every tool on the market scores text with a model. A model's score drifts between runs, so it can advise but cannot block anything. Our checks are ordinary functions over an artifact: the same input gives the same answer.
Each was born from a real fix on a shipped article: an uncited figure, a broken internal link, a promise in the heading the body never keeps.
Measured on the served page, not the build: time to the main element, layout shifts, blocked main thread. A slow page never makes it into a citation.
Every figure carries a type: measured, estimated, proxy, unknown. A gate computes on measured only — a missing source yields «unknown», never an average.
AI crawlers do not run JavaScript. The page has to deliver its text from the server — and that is checked, not assumed.
In Google's and Gemini's answers YouTube outweighs almost everything else. No SEO tool has a video pipeline of its own — for us it is the main product, so one piece of research feeds both the article and the clip.
In practice a topic is worked once and ships twice: as a page that gets cited, and as a clip that brings views and that very link.
The same line that runs clients' social channels — how it works.
Not by a feeling that «we are more visible». By data you can open yourself.
Search Console and Bing Webmaster are connected to your domain and stay yours. Positions, impressions, real queries — from there, for free.
We separately test whether the engines answer your questions with your material, and show where you are cited and where a competitor is.
The data layer runs on single-digit dollars a month, not on an enterprise seat. We do not resell you somebody else's subscription.
We started from a domain with one referring domain and zero organic keywords, against competitors with thousands. That is exactly why the bet is on answer engines rather than on a fight for a place in the results: at our starting point it was the only winnable one.
Everything above ran on our own blog first — mistakes included. The most expensive one: search-box suggestions produce a «relevance» number that is very convenient to sort topics by; checking it against real demand showed a 36× discrepancy. Which is why proxy signals are permanently labelled proxy here and never allowed into a gate.
What comes out of it is visible on our own blog.
None of these can be verified, so we do not offer them.
Rankings
We do not control ranking. We control the structure of the piece, page speed, indexation, and that every figure in the text is backed by a source.
That you will be cited by a date
Indexation in Bing takes days; accumulating trust takes months. We show movement on measurable steps rather than drawing a line going up.
Hundreds of articles a month
Every piece goes through your approval of the brief and through a blocking audit. That caps the pace deliberately: a hundred empty pages hurt a domain rather than help it.
That someone else's data will pass as yours
The material stands on your substance — figures, cases, screenshots. Without it the brief does not pass: an article with no evidence of its own gets cited by nobody.
No. We work with your site too: the material is handed over with markup for your CMS. On ours the only advantage is that the speed and schema audits are already wired in.
Most start with two to four: enough to get the first measurements and see which topics reach the answers at all. Volume grows from the data, not from the budget.
Access to your domain's consoles, an hour to approve the first briefs, and your substance: figures, cases, the objections customers actually raise. You do not write.
Quoted per volume and per whose domain it is: it depends on the number of articles, on whether each needs a video, and on whether we publish or hand over. We name the figure on the call, with the first month's plan.
Then the first step is an audit of the pages you have. It usually turns up indexation closed to AI crawlers, slow pages, and articles without a single checkable fact — cheaper to fix than to write new ones on top.
Tell us the domain and the niche — on the call we show what the engines answer on your topics today, and in whose words.