For people who do this for a living
If you run an agency, you have read a dozen pages of six-pillar frameworks by now. This isn’t one. Below is the work item by item, the evidence grade behind each, and the popular deliverables we turn down. Judge it on the mechanics.
Every claim on this site traces to one number: how many times an assistant names a business. We pull it from Ahrefs across ChatGPT, Perplexity, Gemini, Copilot, Grok and Google AI, we baseline it before we touch anything, and we date it.
Here is what that looked like for three real businesses on the day they started with us, before any work. We don’t name them until they tell us we can.
Ahrefs, July 2026. That gap is the entire job.
Ten items. Each one carries what we know about it and how well we know it, including the one we openly can’t prove.
Google Business Profile, Bing Places, Foursquare, Apple Business Connect, Yelp, and the vertical directories that matter in your client’s trade.
Evidence: These are the retrieval databases themselves. It is a binary gate: absent means invisible, no matter how good the site is.
Software, with owner approval where a platform demands verification
Post-visit asks, and replies drafted for approval. Review text teaches the model what a business is best at, not just that people like it.
Evidence: A core input across every engine. 73% of consumers only trust reviews under 30 days old, so recency is its own lever.
Owner approves the ask-flow once, then software runs it
We can see which third-party lists the assistants pull from, then pursue inclusion on those specific ones.
Evidence: Listicles are 43.8% of the page types ChatGPT cites. This is the highest-evidence local lever and no software on the market does it.
Software finds and drafts, a human sends
Per-service and per-city pages, answer blocks in visible HTML, and, the part most people skip, statistics, quotations and cited sources inside the copy.
Evidence: +27–41% visibility in a controlled experiment (peer-reviewed GEO paper), corroborated by Ahrefs’ 1.4M-prompt study. The effect is largest for lower-ranked sites, which is exactly the local client.
Software writes, owner approves before anything publishes
Scheduled, genuine re-updates of the pages and lists that carry citations.
Evidence: 79% of AI-cited lists were updated within the year; Perplexity’s index decays in days. The citation multiple is roughly 3.2x, vendor-grade, not peer-reviewed.
Software drafts, owner approves the publish
Comparison and “best” content for questions in the client’s market that nobody has claimed yet.
Evidence: Ahrefs ran a field experiment from 0% to 65% mention share in an unclaimed niche. 28% of cited pages have zero organic visibility. You do not need to rank to be cited.
Software drafts, owner approves
City magazines, local news, and journalist requests.
Evidence: Brand mentions correlate 0.66–0.71 with citation. Correlation, and we say so.
Software drafts, a human pitches
True credentials and statistics published in the places assistants quote from: vertical profiles, manufacturer locators, fact-dense social bios, one honest press release.
Evidence: Observed live in a July 2026 capture: ChatGPT quoted Instagram posts, LinkedIn bios and a 2014 press release verbatim while ranking practitioners. Press releases get cited for a decade.
Software drafts, owner approves. We never inflate a credential
IndexNow, Bing Webmaster, and sitemap hygiene.
Evidence: An accelerator, not a driver. It matters because ChatGPT reads Bing, so the difference is hours instead of weeks.
Software, end to end
We own an automated video pipeline, so testing this costs us almost nothing.
Evidence: The strongest single correlate we have found, at 0.737, but causality is unproven and we will not pretend otherwise.
Software produces, owner approves the publish
This is the part worth your attention, because most of it is on sale elsewhere right now.
No engine reads it. Google has confirmed this, and 97% of the files that exist receive zero AI-bot traffic. We keep one on our own site because it costs nothing. Selling it as a deliverable would be selling a file nobody opens.
It is the headline pillar on most AI-SEO service pages. The best evidence available, a study of 1,885 cited pages, found it is a causal null for whether an assistant cites you, because engines extract the visible HTML at retrieval time. We implement it anyway for Bing Copilot and Google rich results. We will not tell you it is the reason you got named.
It gets detected, it gets accounts banned, and it is an FTC problem. Perplexity’s Reddit citations already collapsed 86% after litigation. It is fragile as well as dishonest.
Mass-generated location pages do not index, do not rank, and put the client’s domain at risk. We build the ones a real customer would search for, and no more.
Coaching customers to include keywords in reviews crosses into fake-engagement policy. We ask for the review. We do not write it.
You can build this. It is roughly ten workstreams, a measurement pipeline across six engines, and a publishing loop that gets an owner to approve changes in about ten minutes a week. We already run it, at $495 a month flat, with one guarantee: in 60 days you'll be showing up in ai answers more than you are today. if not, every dollar back.
If you would rather point your local clients at something that already works than staff it yourself, that is the conversation. We don’t touch your paid, your web builds, or your classic search work. This is the AI-answer layer and nothing else.