Answer Engine Optimization Agency: Costs and 9 Questions

Hire an answer engine optimization (AEO) agency only if it will show a before-and-after on the same buyer prompts, in writing, before you pay for a second month. If it won't, work in-house or buy a bounded audit. AEO is also called generative engine optimization (GEO) or "AI SEO"; Wikipedia lists them as names for one field. We are an audit tool, not an agency, so we have no retainer to sell you.

Do you need an AEO agency, a consultant, or neither?

Under roughly $3,500 a month, do it in-house and buy at most a one-off audit; the cheapest published hands-on retainer we found starts at $3,495. A consultant fits when you can implement but need diagnosis. An agency fits when you lack people to produce and ship the work.

Google's own hiring advice says a small local business can probably do much of the work itself, and that a redesign or launch is a great time to hire (Google Search Central). We run our own visibility in-house: our robots.txt allows GPTBot, ClaudeBot and PerplexityBot, and on 2026-10-03 we merged 165 pages into 42, because one strong page per question beats five thin ones. Selling the service instead? See our playbook for agencies.

What should an AEO or GEO engagement actually deliver?

A real engagement delivers six inspectable things: a baseline, a prompt map, access fixes, content changes, off-site corroboration and a retest. Each needs an owner, a target prompt and a way to verify it.

1. Baseline. Run 25 to 50 buyer prompts on each engine you care about. Keep raw dated answers, citations and whether live search was on. Separate mentioned, cited and recommended; they are different outcomes.
2. Prompt map. Group prompts by buyer, stage and objection. "Best payroll software for a 20-person nonprofit" earns work; "what is payroll" does not.
3. Access fixes. Crawlers allowed, pages indexable, bot protection not blocking. OpenAI says sites opted out of OAI-SearchBot are not shown in ChatGPT search answers (OpenAI).
4. Evidence and content. Fix contradictory product claims first, then write pages that settle objections: pricing, comparisons, limits. The original GEO paper reported visibility gains of up to 40%, varying by domain (arXiv): a research result, not a client guarantee.
5. Off-site corroboration. Reviews, profiles, comparisons and research that independent sources can cite. In its US software results, a web-enabled GPT drew 72.7% of software-product sources from earned media against 45.4% for Google; the authors call the percentages illustrative (Chen et al.).
6. Retest. Same prompts, same method, a change log, and uncertainty stated.

Photo by Christina Morillo on Pexels

Schema and llms.txt are missing from that list on purpose. Google says there is no special schema.org markup to add for its generative AI features, and that Google Search doesn't use llms.txt (Google). Do both cheaply; never pay retainer money for them.

What counts as proof that it worked?

Proof is the same prompts, run the same way before and after, reported as how often your brand appears across repeated runs. A single screenshot or an "AI ranking position" is not proof.

SparkToro's 2,961 runs found a below 1-in-100 chance of getting the same brand list twice. Its author judged visibility percentage across many runs reasonable and called ranking-position tracking "baloney" (SparkToro). It was run with Gumshoe.ai, an AI-visibility vendor.

Presence is the number to watch. In AEOeye's recommendation benchmark (127 audits, 892 answers), the audited brand was named in 157 answers (17.6%), and when ranked it came first in 71.4% of 147 cases (data). Being named at all is the hard part.

Caveat: self-selected brands and unequal engine samples, so not a market rate. Our methodology template specifies a protocol; measuring AI visibility covers metrics.

How much do AEO agencies and consultants cost?

Prices are quote-based, and published ranges come from agencies selling the service, not independent surveys. As of October 2026, hands-on programs we found published run from about $3,500 to $14,000 a month; audits and advice cost less.

The guides disagree: treat them as sales-side anchors. Cost drivers are engine and market count, prompt-set size, content volume, paid off-site work and engineering. Ask for month-to-month or 90-day terms; Rock The Rankings says a 90-day commitment saves roughly 10%.

Which 9 questions should you ask before signing?

Write down the answers. A good agency answers each with an artifact (a sample prompt set, a raw-answer record, a brief), not a reassurance.

1. Which buyer questions first, and why? Expect 10 to 25 prompts tied to purchases, not a keyword dump.
2. Which engines, reported how? Separately, with model, date, market and search on or off, never one blended "AI score".
3. How will we set a baseline someone else could rerun? Fixed prompts, repeated runs, written classification rules.
4. What will you change on our site, who writes it, and what won't you publish? Ask for three samples made under contract; refuse fifty generic articles.
5. How do you earn corroboration off our domain? Legitimate reviews, profiles and research; no fake reviews, undisclosed paid lists or mass PR.
6. What do we hold at day 30? Baseline, prompt map, source gaps, access findings, a backlog with owners.
7. How do you handle schema and technical work? Markup must match visible content; someone validates it after template changes.
8. How will you measure progress without invented attribution? Same prompts, a change log, negative results kept, exportable data.
9. What happens if we cancel, and when would you tell us not to hire you? We keep the content, schema and prompt records. Real answers name unclear positioning, no implementer or unstable product claims.

What are the red flags?

Guaranteed placements, a proprietary score with no inputs, and a retainer whose only output is monitoring each end the conversation. Google says no one can guarantee a #1 ranking, and its hiring page asks whether an agency's AEO or GEO advice matches Google's official guidance.

• Guaranteed spots in ChatGPT, Gemini or an AI Overview.
• Screenshots without a date, prompt and engine.
• A prompt set that changes between baseline and follow-up, or hidden competitor results.
• No distinction between model memory and live retrieval.
• Schema, llms.txt or content "chunking" sold as the fix; Google says none is required.
• Placements and community posts with no disclosure. Google says seeking inauthentic mentions isn't as helpful as it might seem.
• Invented statistics or quotes; "training" ChatGPT with a few articles.
• Write access to Search Console during an audit; Google advises read access only.

Photo by RDNE Stock project on Pexels

How should a 90-day pilot be structured?

A pilot is one audience, market and product line for 90 days, at a fixed price, with a decision date set before kickoff. It ends with a before-and-after on the frozen prompt set. Expand only if it produced shipped changes and measured movement.

• Days 1 to 30: baseline, frozen prompts, a ranked problem list.
• Days 31 to 60: ship owned changes: access fixes, claim corrections, the few pages that settle real objections.
• Days 61 to 90: retest identically, compare, and write down what did not work.

Agree in writing: engines, prompts, competitors, target pages and claims, owners, success and stop criteria, and who owns the assets. Our AEO audit checklist and AEO strategy cover what to test and decide.

Before you brief anyone, get your own baseline. The free audit asks ChatGPT three unbranded buyer questions and shows full answers and the competitors named instead. The $29 one-time report adds Claude, Gemini, Google AI and Perplexity and 10 questions, each engine at the default tier a free user sees. If an agency's baseline contradicts yours, ask why.