TL;DR
- Journalism alone makes up 27% of the sources cited in ChatGPT, Claude and Gemini answers, and paid and advertorial content 0.3%, in Muck Rack's May 2026 study of more than 25 million links across 17 industries. A service that works only on pages you own leaves much of what AI cites untouched, so settle first who handles the rest.
- As our AI visibility hub defines them, GEO (Generative Engine Optimization) targets LLM-powered platforms like ChatGPT and Perplexity, while AEO (Answer Engine Optimization) is structuring a page so an answer engine, from Google's featured snippets and voice assistants to ChatGPT, can lift a direct answer from it.
- The engines document less than a proposal implies: crawler access, the answer near the top of the page (Bing's rule, not Google's) and consistent naming. Structured data they treat as optional and bound to what the page shows, and the prompt set every tracker relies on is the seller's method, not an engine's. Google's guide says to ignore llms.txt files, chunking and AI-specific rewrites for Google Search.
- Google says you cannot mark a page as a featured snippet and Bing says GEO does not guarantee citations, so a guaranteed citation is a reason to walk away. Google's hiring page now tells buyers to check AEO and GEO advice against its documentation.
What do AI answers cite, and where does that leave AEO services?
Much of what AI answers cite is not on your website. For its May 2026 edition, Muck Rack's Generative Pulse team analyzed more than 25 million links from ChatGPT, Claude and Gemini responses across 17 industries: journalism alone made up 27% of cited sources, and paid and advertorial content accounted for 0.3%. A controlled study across multiple verticals, languages and query paraphrases found that AI search shows a "systematic and overwhelming bias towards Earned media" over brand-owned and social content.
That is the PR half of AI visibility, and it is the half an AEO scope leaves out. AEO services work on the other half, the pages you own: whether a page can be reached, understood and quoted when an engine looks at your site, and whether its facts match what everyone else says about you. That work can raise the odds that an engine selects a page, which is what Bing says of URLs focused on a single topic, but it cannot make it.
The research on page changes says less than a sales deck may suggest. The paper that introduced generative engine optimization reported gains of up to 40% on its own benchmark and said the efficacy of its methods "varies across domains". A 2026 study of 602 controlled prompts describes the pages that get cited without proving that imitating them earns a citation. If either turns up in a proposal, ask which domains and which engines.
So the first question for any AEO proposal is what it does about the sources you do not own, and who does that work. Our guide to PR for SEO covers that side.
What should an AEO agency deliver?
A credible scope names what you will hold in hand and says where each requirement comes from: an engine's documentation, the seller's own testing, or plain project hygiene. We would expect these six.
Question research from real buyer prompts
What you receive: the questions your buyers put to AI engines about your category, each mapped to the page that should answer it or to a gap where none does. The map stops a scope from becoming a page for every phrasing, which Google warns against: creating separate content for every variation of how people might search, fan-out queries included, when done "primarily to manipulate rankings or generative AI responses in Google Search violates Google's scaled content abuse spam policy".
Answer-first revisions to priority pages
What you receive: a short list of pages, by URL, with the copy before and after. Each revised page covers one primary topic, with the answer near the top and its facts stated outright. Those are Bing's rules for Copilot and grounding: "URLs focused on a primary topic are more likely to be selected for grounding results". Google's guide asks for none of it for AI Overviews or AI Mode, so a proposal that credits Google with the answer-first rule has the source wrong. Our post on answer-first page structure has the how-to; our position is that a site-wide rewrite sold before the diagnosis is the wrong order.
Entity and fact consistency
What you receive: one canonical description of the company and each product, used on the site, in your Organization structured data, and handed to everyone who writes about you. Bing asks for clear and consistent naming and says "Clear entity definition improves grounding visibility and citation accuracy". We have seen why. In two answers we read on 4 October 2026 while tracking the prompts behind this post in Gauge, the AI visibility measurement platform we use with clients, Perplexity did not connect the bare acronym AEO to this work at all: one answer described an unrelated organization, and the other said there was no widely recognized term. Two answers are an observation, not a rate, but an engine that cannot pin a name to your company has nothing to attach your facts to.
Crawler access, tested per crawler
What you receive: a recorded result for each crawler that matters to you (Googlebot, Bingbot, OAI-SearchBot, PerplexityBot, and Anthropic's two, Claude-SearchBot for search indexing and Claude-User for fetching a page when someone asks Claude a question), with the cause of each failure and its owner. Google lists "Ensuring that crawling is allowed in robots.txt, and by any CDN or hosting infrastructure" among its fundamentals; OpenAI asks you to allow requests from its published IP ranges; Perplexity says that with a web application firewall "you may need to explicitly whitelist Perplexity's bots to ensure they can access your content". The test should also confirm that the Search generative AI control in Search Console still reads "Include my site's links and content in Search generative AI features", the default, since Google says a site must be included there to be eligible.
Structured data that matches the page
What you receive: a validated inventory of the markup on the pages in scope, with anything that describes content a reader cannot see removed. Google: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add". Bing: structured data "may support clearer grounding but does not guarantee visibility or grounding traffic".
Measurement you can re-run
What you receive: a baseline taken before any page changes, a prompt set that stays the same between re-runs with any refresh written down, and the first-party reports connected in your name: Search Console, Bing Webmaster Tools and your analytics. You should own the prompt set, the exports and the account access, or the history leaves with the agency.
What do Google, Bing, OpenAI, Perplexity and Anthropic document?
Each vendor's words apply to its own products; nothing Google says tells you how ChatGPT, Claude or Perplexity choose sources. The deliverables above cover what each asks of a page. This is what each says about guarantees and limits.
Two of the five, Google and Bing, publish page-level guidance. Google ties its AI features to ordinary search, "rooted in our core Search ranking and quality systems". The other three publish crawler controls. The one markup note we can find from them is OpenAI's, that its ChatGPT Atlas browser agent "uses ARIA tags", and that is guidance for an agent clicking through a page, not for answer formatting. So a claim about the page format ChatGPT, Claude or Perplexity prefers rests on the seller's own testing or inference, and a seller should say which.
Neither engine that publishes on the question will hand anyone the answer slot, and we have found nothing from the other three that reads as a promise. Google's featured snippet documentation asks "How can I mark my page as a featured snippet?" and answers: "You can't. Google systems determine whether a page would make a good featured snippet for a user's search request, and if so, elevates it".
Our post on whether your site blocks AI crawlers covers robots.txt and server-level permissions; in our experience the CDN and firewall layer is the part most easily skipped.
Which deliverables do the engines say you don't need?
Read a proposal's line items against Google's list of what you can ignore for Google Search: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search"; "You don't need to write in a specific way just for generative AI search"; and seeking inauthentic mentions "isn't as helpful as it might seem". Line items to question:
- An llms.txt file as a deliverable. Google's guide says keeping one for other services is fine and that doing so "will neither harm nor help your site's visibility or rankings in Google Search, as Google Search ignores them", a point it added on 15 June 2026. A scope built around it is a file in place of work.
- FAQ schema sold for rich results. Google retired the FAQ rich result on 7 May 2026. Markup sold as a route to that result is selling a feature that no longer exists.
- Chunked, AI-rewritten or mass-produced copy. None of it has a documented benefit in Google Search, and Bing's abuse list goes further: "Content should be created for users, not to manipulate ranking systems or trigger citations or AI responses", and large-scale content produced without editorial review "may be excluded from indexing".
- Text written to steer the model. Bing lists prompt injection under abuse and says content designed to manipulate or interfere with its language models "may result in reduced visibility or removal from search experiences".
- An AEO score with no method. Google says "No third-party tool has access to our internal ranking or AI systems". That covers every prompt tracker, ours included: a score is a sample of answers, and the proposal should say how it is built.
How do AEO services differ from an SEO retainer, and from GEO?
Much of the foundation is the same work, and Google says so: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO".
What an AEO scope adds, per the documentation, is specific: AI search crawlers tested through the firewall and CDN; the Search generative AI control confirmed; answer-first restructuring of the pages that should answer buyer questions; entity consistency; the Google and Bing AI reports connected; and a prompt set tracked over time. A renamed SEO retainer may contain none of these, which is why the scope has to name them rather than the label. Our comparison of SEO, AEO and GEO lays out the overlap.
GEO is the wider purchase. As that comparison puts it, "SEO and AEO are work you can do on pages you control. GEO adds every page you do not", which is why it pulls PR into the same program. Buyers and AI answers often use the two names for the same purchase, so ask what the scope covers off your site instead of trusting the label. The off-site work, the first 90 days and the red flags are in our guide to what a GEO agency actually does.
How should results be measured?
A credible report separates work completed, engine-reported impressions and citations, referral visits with their key events, and sampled change across a prompt set.
Google's Generative AI performance report counts impressions, defined as "how many times links to your site were shown to a user in a generative AI feature on Google Search", for AI Overviews and AI Mode, by date, page, country and device. The help page lists no query view and no click metric, and the report covers Google's features only. Google says it rolled the report out to all websites worldwide as of 31 August 2026, though the same page still notes that not every property has it.
Bing's AI Performance report shows which pages were cited and the grounding queries behind them, across Microsoft Copilot, AI summaries in Bing and partner integrations. Bing says the report "does not measure rankings, authority, performance, or importance", that the data "represents a sample of overall citation activity", and that trends "cannot be attributed to a specific cause or event".
ChatGPT referrals are where you see behavior. OpenAI says "ChatGPT automatically includes the UTM parameter utm_source=chatgpt.com in referral URLs", so you can isolate those sessions and see which key events they complete. Referrals never show an answer nobody clicked.
A prompt set re-run on a schedule fills the gap. Gauge, the AI visibility measurement platform we use with clients, samples ChatGPT, Perplexity, Gemini, Copilot, Google AI Overviews and AI Mode; it does not sample Claude. Any tracker is a sample: a comparison of five generative search systems found their outputs "can vary across time and executions", so a trend needs several runs of the same prompts. No tracker, ours included, can prove a heading rewrite caused an answer to change.
What should you ask in the pitch, and what belongs in the statement of work?
The pitch tests whether a scope is specific; the statement of work is where that has to survive. Google's hiring page, "Do you need an SEO?", lists "Optimizing for generative AI" among useful agency services, notes it is also sold as AEO and GEO services, and tells you to ask whether it is "aligned with Google Search's official guidance on optimizing for generative AI features".
Questions to ask
- Which of our pages would you change first, and what evidence supports that choice?
- Which search and AI crawlers will you test, and how will you test access through our firewall and CDN, not only robots.txt?
- What will you do about the earned media and third-party sources we do not own, and who does that work?
- Which engines will you measure, how many prompts, how often re-run, and who owns the data?
- Who will do the audit, rewrite the pages, test the crawlers, implement the changes and write the report?
Google adds a test for the meeting: "Be careful if a company is secretive or won't clearly explain what they intend to do".
What the statement of work should name
A useful statement of work replaces promises with named assets, owners and outputs: the pages in scope by URL; the crawlers to be tested and who fixes what the test finds; the canonical company and product descriptions; the structured data types per page; the prompt set, how often it is re-run, when it is refreshed, which engines it covers and the reports you receive; who approves and who implements each page change; and the access level at each stage (Google advises read-only Search Console access for the audit). It should say who owns the data and accounts when the engagement ends. It should guarantee nothing about a snippet or a citation. Google's rule transfers: "If they guarantee you that their changes will give you first place in search results, find someone else".
What drives the cost, and can you do it in-house?
The cost of the on-site work follows the number of pages that need restructuring, how much technical debt the crawler test turns up, whether the agency implements changes or only recommends them, how many engines and prompts are tracked and how often, and whether earned media is inside the scope at all. In our experience the last moves the cost most, because placements on sites you do not own are PR work on a different clock.
Much of the on-site half is doable in-house: a content lead can run the question research and restructure pages, a developer can test crawler access and fix what the test finds, and Search Console, Bing Webmaster Tools and your analytics supply the first-party reports. In our experience, what an in-house team runs short of is a prompt schedule that is kept to, and the earned media that sits outside AEO altogether. Our GEO agency guide covers the in-house decision in full. Althea Labs prices each engagement after a scoping call.
How does Althea Labs approach this work?
Althea Labs is a brand presence and AI visibility agency that builds targeted visibility systems for B2B startups through integrated PR, SEO, and content programs. Althea Labs covers GEO and AEO under a single visibility program. Presence beats fame. On who does the work: you work directly with Kara Silverman, Althea Labs' founder, throughout.
AI visibility is the core of the work. We start on the pages you own: the pages that should answer buyer questions, their structure, crawler access, and structured data that matches what the page shows. We then work on the content off your site that AI tools cite, and we measure both against a baseline, with the same prompts re-run on a schedule. PR, the pitching that earns media placements, is a premium add-on that we scope separately. Our services page lists the packages.
An engagement runs in five steps, in order.
- Audit. We review your SEO content and technical SEO, analyze your PR footprint across owned and earned media, baseline your AI visibility and entity recognition, and deliver a Brand Presence Report with prioritized recommendations.
- ICP + Messaging. We define your buyer and their buying triggers, and build the messaging framework.
- Visibility Tracking. We set up Gauge and baseline 300 to 500 prompts pulled from real conversations and your own customer interactions, then re-run a tracked set every month.
- Strategy. We turn the findings into a 90-day editorial calendar across content and community, and PR where it is in scope.
- Execution. We do the monthly work: writing, PR pitching where PR is in scope, social media, paid opportunities, speaking submissions and award nominations, with a monthly visibility report against the baseline.
For a narrower project, our AI Comms Toolkit includes up to three website page structures for answer engine optimization.
FAQs
Are AEO services and GEO services the same purchase? No, though sellers use the labels loosely. AEO is the on-site job, structuring a page so an answer engine can lift a direct answer from it; GEO targets LLM-powered platforms like ChatGPT and Perplexity and, as our comparison guide has it, adds every page you do not control. Our AI visibility hub has the full definitions.
Can an AEO agency guarantee a featured snippet or a citation? No. Google says you cannot mark your page as a featured snippet, and Bing says GEO does not guarantee grounding or citations in AI experiences. Google's hiring page adds "No one can guarantee a #1 ranking on Google", and we put a guaranteed citation in the same category.
Do I need AEO if my SEO is already strong? Strong SEO is most of the foundation, because Google says its generative features are rooted in its core Search systems. AEO adds crawler tests through the firewall, answer-first structure on priority pages, consistent entity facts, the Google and Bing AI reports, and a tracked prompt set. If your retainer covers those, you do not need a second one; if not, ask for them by name.
Is FAQ schema still worth adding? Only as hygiene. Google stopped showing FAQ rich results on 7 May 2026, so FAQPage markup no longer earns that result. Our position on the markup is in the page-structure post linked above. The visible question-and-answer section is worth keeping either way: Bing's AI Performance guidance names FAQ-style content among the formats that make information easier to reference in its answers.
How long before AEO work shows up in AI answers? Take a baseline in the first month, before any page changes. In our experience, early signs such as a cleaner site and first placements show up within 60 to 90 days of consistent work, and movement in AI answers builds on those inputs over the months that follow. OpenAI and Perplexity both say a crawler-access change alone can take around 24 hours to register, so judge by the same prompts re-run on a schedule, never by one answer.
Get an audit before you buy a retainer
If a proposal cannot name the visibility problem the retainer will fix, the audit comes first. Our Brand Intelligence Audit is delivered in five days as a deck plus a one-hour session and covers your SEO footprint, your earned media, your AI visibility and your community presence, so every proposal you receive afterwards is held against the same facts. Start your audit before you sign for ongoing execution.
CONTINUE READING
