Getting named by AI: the honest guide to GEO tools and agencies

What GEO is, in one paragraph

Generative engine optimization is the work of being the name inside AI answers, not a blue link under them. The measurable unit is visibility: across a fixed set of buyer questions, asked repeatedly, in what share of answers are you named? Measured that way, over dozens to hundreds of runs, it is a stable, real metric (SparkToro found brand visibility stable within narrow bands across nearly a thousand runs). Measured as a one-off screenshot, it is noise: the same question rarely returns the same brand list twice.

The monitoring tools, compared

Every tool below sells the same core loop: fan a prompt set out across AI engines, parse brand mentions, compute visibility and share of voice, analyze which sources the engines cite. Prices are as reported in July 2026; confirm current pricing with each vendor.

The uncomfortable truth about this market: the core monitoring loop is commodity engineering. What you are really paying the premium vendors for is browser-based capture of consumer surfaces (API answers are not always what users see) and real-user prompt datasets. If a cheap tool and an expensive tool disagree, that is usually why.

What actually moves the number (with evidence)

• Statistics, quotations, and cited sources in your content. The Princeton GEO study (arXiv 2311.09735) measured roughly 30-40% visibility lift from exactly this. Simulated engines, so treat it as directional, but it matches what we see on our own domain.
• Third-party presence. ~85% of AI brand mentions come from pages you do not own: comparisons, review sites, communities, directories. Two or three honest third-party mentions usually beat anything you publish on your own domain.
• Directly answering the questions buyers ask, one page per question, kept fresh. Engines assemble answers from citable sources; be the citable source.
• Measuring as an aggregate trend. Fixed prompt set, repeated runs, visibility % over time. It is the only defensible way to know if any of this is working.

The snake oil list

• "AI rank position" screenshots. Engines rarely return the same list twice. Any vendor showing you a single-run rank is selling variance as insight.
• Keyword stuffing for engines. Measured effect in the same Princeton study: negative. It reads as spam to the models the same way it reads to humans.
• Betting on one platform's citation habits. ChatGPT's Reddit citations collapsed from roughly 60% to 10% of certain answer types in late 2025 after one retrieval change. Seeding a single platform is renting, not building.
• llms.txt as a magic ticket. Ahrefs' 137,000-domain log study found ~97% of llms.txt files are never fetched by anything. We publish one anyway, because it costs an hour and the agents that do read it get exactly what we want them to know, but anyone selling llms.txt as the strategy is selling you a file.

Tool, agency, or agent: who should buy what

Buy a monitoring tool if you have a content team with spare capacity. The tool tells you where you stand; your team does the work. Cheapest path if the capacity is real.

Hire an agency if you want the work done and are comfortable with a monthly retainer. Ask exactly two diligence questions: show me visibility measured as a trend across a fixed prompt set, and show me what you shipped last month for a client.

An always-on agent is the newest option and the one we sell, so apply discount accordingly: software that measures daily and does the work daily, with a human approving anything public. Our version is Lookout; it runs on this domain and the chart is public, which is also our answer to the case-study problem: as of July 2026 nobody in this market, including us, has a widely verified study tying AI share-of-voice to B2B revenue. Until someone does, prefer vendors who show you live measurement over vendors who show you promises.