AEO (answer engine optimization) is the work of getting a brand named and cited (shown as a linked source) in ChatGPT, Perplexity and Google's AI answers. It can be bought two ways, and the two sell different things. Software sells an instrument: tools that track where you stand in AI answers and, in the fuller versions, plan, write and publish content and chase mentions. An agency sells labor and judgment: people who do the weekly work, plus someone senior who decides which work matters. Which fits mostly comes down to one question: who does the work every week.
On price, RankControl collected 14 agency price lists in October 2026. Specialist retainers in them ran from low four figures to five figures a month, with enterprise programs above that. Freelancers on hiring boards list AEO and GEO work from a few tens of dollars an hour, with a quality spread to match. Monitoring software starts at about $99 a month, and a full pipeline (software that plans, writes, publishes and tracks) costs about $400 a month.
Each option has a hidden cost. An agency's is opacity and dependency: what the program learns builds up in the agency's heads, and the client's data sits in its reports. Software's is hours: a tool without an operator is just a subscription, and the fuller pipelines reduce the weekly hours without removing them.
The market is already splitting the job. Many agencies run client programs on the same software a buyer could purchase, so the doing is turning into a commodity (cheap and interchangeable), vendors keep attaching human strategists to their software, and freelance boards price AEO work at commodity hourly rates. What is scarce is judgment and evidence, so a hybrid is spreading: software runs the weekly loop on cost and consistency, and a human sits on top for judgment, either as an optional strategist added to a pipeline or as the buyer's own fractional or in-house operator.
A made-up SaaS company run through both options for 13 weeks shows where the learning ends up. With an agency, weeks 1 and 2 go on onboarding calls and a strategy deck, work starts in week 3, the week-6 monthly report shows the first baselines of how often AI answers cite the company, some real wins arrive by week 10, and the week-13 quarterly review is where the client learns what happened. With software, tracking goes live in week 1, articles ship by week 2, the lines for each AI engine react by week 5, and an operator makes steering calls every Monday. By week 13 the progress is comparable; the learning stays mostly with the agency in one case and in-house in the other.
An agency fits a company with no internal marketing capacity, a real strategy vacuum, or an enterprise with multiple brands, procurement, legal review and a dozen stakeholders. Software fits a team with an operator who can own the weekly loop of tracking citations, shipping content and earning mentions.
Vetting an agency takes three checks: two or three case studies from businesses like yours, with before-and-after numbers and a call with one of those clients; directories for the shortlist only; and whether the firm shows up when ChatGPT and Perplexity are asked about its own niche. Five questions for the call. Which tools will the agency run the program on, and can the client see the data in them directly? Who works the account, and how many accounts does that person carry? Can the client see a company like theirs, before and after, and call it? What happens to the client's data, prompts and learnings if the two part ways? What did the agency ship for its last client in the first thirty days?
Four questions for a software call. Can every number be traced to an actual answer and cited URL? Is the prompt panel, the fixed list of buyer questions the tool runs, really fixed, and what is the cadence? What does the tool do, and what does it leave to the buyer? Can the buyer see articles it published for a real customer, on that customer's own domain?
The failure modes are not symmetric. Choosing the wrong agency costs time and opacity: two quarters and a five-figure sum before the pattern is undeniable, then a restart for a program whose memory lived in someone else's head. The wrong software costs quieter money. It is a subscription nobody operates, and it is cheap enough to run unexamined for a year. A 90-day review with the evidence on the table, booked before signing, covers both.
Ram writes about this for RankControl, an AI SEO and AEO platform. RankControl's AI visibility tracker tracks citations across 7 AI engines: ChatGPT, Perplexity, Claude, Gemini, Grok, Google AI Mode and Google AI Overviews. The full write-up has the vetting checks and the week-by-week quarter.