The short answer
Choose the SEO agency that can show you, engine by engine, which of your buyers' questions name you today, who gets named instead, and exactly what it will change on your site and off it to move those numbers. Then hold it to the same measures every month. Walk away from any agency that guarantees a spot in ChatGPT, sells a vague "LLM integration", or reports activity instead of outcomes.
Answer engine optimization (AEO) and AI visibility work are real, but the market is young, so the words on agency sales pages run ahead of what can be measured. This guide gives you the questions to ask, the red flags, the results to demand, the homepage schema a competent agency ships early, the contract terms that protect you, and an honest view of when software covers the work for less.
What AEO and AI visibility work actually involves
Before you compare proposals, know what the work is. Answer engine optimization means getting your content chosen as the answer: in featured snippets, People Also Ask, voice results, and the answers that ChatGPT, Google AI Overviews, Gemini, Perplexity and Claude write. Almost all of it builds on ordinary SEO, because most AI engines search the web before they answer. A page that cannot be crawled, indexed and trusted in classic search is rarely quoted.
A serious AEO engagement covers five areas. If a proposal skips one, ask why.
- Access. AI crawlers such as GPTBot, OAI-SearchBot, PerplexityBot and ClaudeBot can reach your pages, your robots.txt does not block them by accident, your important text is in the HTML rather than loaded later by JavaScript, and your pages are indexed in both Google and Bing.
- Answer-ready pages. Each key page answers one real question near the top, in plain sentences a model can lift and quote, with specific facts rather than slogans.
- Structured data. Schema that tells machines who you are, what you sell and who wrote what, matching the visible page.
- Off-site presence. AI answers often quote review platforms, comparison articles, directories, community threads and reference sites rather than your own pages. Getting you into the sources engines already cite is a large part of the work.
- Measurement. A fixed set of questions asked on each engine, repeatedly, so you can see the trend rather than one lucky answer.
For the full step-by-step version of the work, see our guide on how to show up in AI search engines. Reading it before agency calls makes it easier to spot a proposal that is all promise and no plan.
How agencies get clients featured in content AI engines cite
When an engine answers "best X for Y", it usually leans on a handful of pages: review sites, roundups, comparison posts, forum threads and the vendors' own pages. Good agencies find those exact pages for your target questions, then work to get you included honestly: accurate listings on review platforms, pitches to the authors of roundups that leave you out, data or expert quotes journalists can use, and useful answers in the communities your buyers read. Ask any agency to show you this source list for three of your questions before you sign. If it cannot, it is not doing AEO yet.
Questions to ask before you sign
Use these in the first call. Good agencies answer them without hesitation and in specifics.
- Which AI engines will you track, and how often? You want named engines (ChatGPT, Google AI Overviews, Gemini, Perplexity, Claude, or the ones your buyers use) and a stated cadence, not "all the major AIs".
- Which questions will you track, and who chooses them? The list should come from your buyers' real questions across the journey, from "what is" to "best tool for" to "X vs Y", and you should approve it.
- How do you handle answers that change from run to run? AI answers vary. A good answer involves repeated runs and trends over weeks, not a screenshot of a single good day.
- Can you show our baseline before we start? Today's numbers per engine, and the competitors named in our place. Without a baseline nobody can prove progress later.
- What exactly will you change in the first 60 days? Ask for a list of pages, fixes and off-site targets, not a methodology slide.
- What off-site work is included, and who does the outreach? Reviews, roundups, PR and community work are often the difference, and often quietly excluded.
- For every content piece, what question does it answer and what outcome do you expect? If nobody can name the target question, the search intent and the competitors already answering it, the piece is guesswork.
- Who owns the work and the accounts? Your content, your code, your Search Console, Bing Webmaster Tools and analytics accounts should stay yours.
- How do you report AI visibility separately from classic rankings? Both matter, but they move differently and should not be blended into one flattering chart.
- Which client can we talk to about AI visibility results specifically, and what was measured? Generic SEO wins from five years ago do not tell you much about this work.
Red flags
- Guaranteed placement in ChatGPT or AI Overviews. No one controls what a model says. An agency can raise your odds; it cannot promise a spot.
- "LLM integration" or "submission to AI engines" with no detail. There is no form that puts your pages into ChatGPT's answers. Ask what the phrase means in concrete tasks. If the answer is a feed, a plugin or a secret partnership, be careful.
- Single-screenshot reporting. One answer where you appear proves nothing, because the next run may differ. Ask for rates across many runs and many questions.
- No baseline, or a baseline that only covers branded questions. Being named when someone types your brand is easy. The commercial value is in unbranded questions like "best X for Y".
- Mass AI-written pages. Hundreds of thin pages published at once tend to hurt classic rankings and give engines nothing new to quote.
- Bought links, fake reviews, or planted forum posts. These break platform rules and can poison the very sources engines read.
- A single file sold as the fix. An llms.txt file or one schema snippet can be part of the work, but neither is a ranking lever on its own, and Google Search ignores llms.txt.
- A retainer with an undefined scope. When nobody brings a data-backed brief, agencies default to their own guesses and report hours spent. That is how budgets leak without anyone noticing.
- Long lock-ins with no exit or handover terms.
What results to demand, and how to read them
Agree the report format before you sign. A useful monthly AI visibility report has these parts:
- Mention and citation rate per engine. Out of your tracked questions, how often each engine names your brand and how often it links to your pages, shown as a trend over months, not a single number.
- Share of voice against named competitors. Who gets named for the same questions, and how that is shifting.
- Which of your pages get cited, and which competitor pages win instead. This tells you what to fix next.
- The sources engines lean on. The third-party pages that keep appearing, and progress on getting you into them.
- A changelog. Every change shipped to your site and every off-site placement, with dates, so movement can be tied to work.
- Classic numbers alongside. Search Console clicks and impressions, and sessions referred by AI platforms such as chatgpt.com, perplexity.ai and gemini.google.com in your analytics.
Read single changes with care: a jump or drop in one run is a reason to check again, not a verdict. Trends across several runs are what count. If you want to understand the metrics well enough to challenge a report, our guide on how AI visibility is measured walks through each one.
The homepage schema an AEO agency should ship
Structured data will not make an engine cite you on its own, but it removes guesswork about who you are. A competent agency gets this right in the first weeks:
- Organization with your legal name, URL, logo, contact point, and
sameAslinks to your official profiles (LinkedIn, review platforms, Wikidata or Crunchbase where they exist). This is what ties your brand to the places engines cross-check. - WebSite with your site name, so engines use the name you want.
- LocalBusiness (or a more specific type) with address and opening hours, if you serve customers in a place.
- Product, SoftwareApplication or Service on the pages that describe what you sell, with prices only if they are shown on the page.
- Article with a named Person author on guides and posts, linking to a real author page.
- FAQPage only where the page really has questions and answers. Google now shows FAQ rich results only for a small set of authoritative sites, so treat it as page description, not a shortcut.
Every value must match what a visitor sees on the page, and the markup should pass Google's Rich Results Test and the Schema Markup Validator. A minimal homepage example:
{
"@context": "https://schema.org",
"@type": "Organization",
"name": "Example Co",
"url": "https://example.com",
"logo": "https://example.com/logo.png",
"sameAs": [
"https://www.linkedin.com/company/example-co",
"https://www.g2.com/products/example-co"
]
}For which types matter most and the mistakes that waste the effort, see which schema types AI search engines use.
Contract terms worth insisting on
- A short first term. Three months, then rolling monthly, with a review against the baseline at day 90.
- A written monthly scope. Pages, fixes, content pieces and off-site targets, so "activity" cannot stand in for delivery.
- Approval before anything ships. Changes to your site and anything published in your name go through you first.
- Ownership. You own all content, code and accounts; the agency gets access, not ownership.
- Access to the raw tracking data. The question list, the answers and the cited sources, exportable, so you can switch agencies without losing history.
- Disclosure of AI-written work and the name of the person who reviews it.
- No payment tied to guaranteed AI placement, because nobody can guarantee it.
- Handover terms for when the engagement ends.
When software is enough, and when you still need an agency
Much of AEO is repeatable: asking the same questions on every engine, spotting who gets named, finding the pages that get cited, fixing crawler access, titles and schema, drafting answer-ready pages and publishing them. Software does that part well and cheaply. What software does not do is judgement and relationships: positioning, digital PR, earning links and mentions, original research, brand, and writing that needs real subject expertise. Those are where a good agency earns its fee.
Raechal sits on the repeatable side, and we would rather say so plainly than pretend it replaces an agency:
- It asks your tracked questions on ChatGPT, Google AI Overviews, Gemini, Perplexity and Claude: one engine on Free, rising to all five on the Agency plan. ChatGPT is asked live, and you see the searches it ran and the pages it read.
- It tracks competitors per site (2 on Free, then 3, 5, 10 and 20 as plans go up) and shows who gets named in your place.
- On paid plans a weekly check re-runs your site audit and your questions and shows what changed.
- It audits and generates schema, and checks whether your brand appears on the review, entity and community sites engines cite.
- Its agents for WordPress, Webflow, Wix and GitHub stage every fix for your approval before it goes live.
List prices run from $39 a month on Starter to $119 on Growth, $249 on Pro and $499 on Agency, so you can see what each plan covers and compare that with the tracking line in an agency proposal.
Plenty of teams do both. They run the tracking and the technical fixes in software, then hand the agency a data-backed brief: the questions they lose, the competitors who win them, the sources engines quote. Most agencies do better work against that kind of clarity, and you stop paying agency rates for the parts a tool repeats every week.
Frequently asked questions
What does an AEO agency actually do?
It makes your content easier for answer engines to find, trust and quote: fixing crawler access and indexing, rewriting key pages to answer questions directly, adding structured data, getting you into the third-party sources engines cite, and tracking how often each engine names you over time.
Can an agency guarantee my business shows up in ChatGPT?
No. Nobody controls what a model writes, and answers change from run to run. An agency can improve the odds and show the trend; a guarantee is a red flag.
How long before AI visibility changes?
For engines that search the web live, such as ChatGPT search and Perplexity, changes to pages they already index can show up within weeks. Anything that depends on what a model learned in training moves far more slowly. Judge an agency on the trend over a few months, not on the first report.
Is AEO different from GEO?
They overlap. AEO is about being chosen as the answer, including featured snippets and People Also Ask. GEO, generative engine optimization, is about being named and cited in AI-written answers. In practice one agency should cover both, because the same pages and sources drive them.
How should I compare an agency's price with software?
Split the proposal into repeatable work (tracking, audits, technical fixes, publishing) and judgement work (strategy, PR, expert content). Price the first part against software, then decide whether the agency's judgement work is worth the rest of the fee.
Should I replace my current SEO agency?
Not necessarily. Bring a data-backed brief to your next meeting: the questions you want to win, who wins them now, and the outcome you expect per piece. Many agencies do better work once they have clear targets.
