The short answer.
An AEO agency finds out what AI engines say when your buyers ask for a supplier like you, then changes the things those engines read until you are named more often and described correctly.
None of this is a secret technique. What separates a good agency from a poor one is whether each job leaves behind something you can open and check. That is the test this guide applies, and it is how we scope our own GEO and AEO services.
- 01 · BASELINETest your buyers’ questions
- 02 · ACCESSLet AI crawlers read the site
- 03 · PAGESAnswer the questions on your pages
- 04 · SOURCESFix what other sites say
- 05 · PROOFRe-test and report the change
If a proposal skips the first step, nothing after it can be measured.
AEO agency or GEO agency: is there a difference?
In practice, no. “AEO agency” and “GEO agency” are two labels for the same service, and many firms use both. AEO stands for answer engine optimisation and GEO for generative engine optimisation. The second term comes from a research paper first posted on 16 November 2023, which introduced “Generative Engine Optimization (GEO)” and reported that its methods could raise a source’s visibility in generated answers “by up to 40%”.
Google treats both as part of ordinary search work. Its guide to generative AI search says: “From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” If you want the three terms set side by side before you speak to anyone, our explainer on how AEO, GEO and SEO overlap does that.
Why buyers are asking for this now
The demand is real even if the labels are loose. G2 surveyed 1,076 B2B software buyers in March 2026 and found that 51% now start their research with an AI chatbot more often than with Google, up from 29% in its 2025 report. In the same survey, 71% said they rely on AI chatbots at some point in their research.
The five jobs a GEO or AEO agency does.
Every credible engagement breaks down into the same five jobs. The table shows what each involves and what it should leave you with.
| Job | What the agency does | What you receive |
|---|---|---|
| 1. Baseline | Agrees a list of real buyer questions with you, runs each one several times in each engine, and records who is named and cited | A dated sheet of questions, engines and raw answers |
| 2. Technical access | Checks robots.txt, firewall rules, indexing and whether your content is in the page’s HTML | A short list of blockers, each with a fix and an owner |
| 3. Pages | Rewrites or creates the pages that should answer the questions you are missing from | Named pages, by URL, with the question each one answers |
| 4. Third-party sources | Finds the review sites, directories, lists and publications engines cite for your category, then corrects and improves your presence there | A source list with the status of each profile or mention |
| 5. Reporting | Re-runs the same questions on a schedule and connects AI referral visits to enquiries in your analytics | A monthly report with the raw answers behind every number |
Baseline: the job everything else depends on
A baseline is a record of what the engines say today, before any work starts. It needs the exact questions, the engine, the date, the country and the full answer. Answers can differ by country, and we cover what is different for UK B2B firms separately. Our walkthrough on testing your own AI search visibility by hand gives the method; an agency should do the same at larger scale, with repeat runs.
Technical access: usually small, sometimes decisive
An engine cannot cite a page it cannot fetch. OpenAI’s documentation says sites opted out of its OAI-SearchBot crawler “will not be shown in ChatGPT search answers”. Such blocks can be accidental, and the agency’s job is to find them, not to sell you new file formats. Our guide to llms.txt, schema and AI crawler access covers what the AI companies’ own documentation says matters.
Pages: where most of the hours should go
Engines quote passages that answer a question directly, so most of the agency’s time should go into your service, comparison and explanation pages. In our study of where AI answers get their sources, we recorded every source cited in 36 answers from Google’s AI Overviews, ChatGPT and Gemini to 12 buying questions on 10 November 2026. About 91% of the 203 citations went to company or vendor websites, which is an upper bound under our counting rules. See our guide to the page patterns AI engines lift and credit.
Third-party sources: slow, and easy to fake
Engines also read what other websites say about you. Honest work here means claiming profiles, correcting old facts and asking real clients for reviews. See our breakdown of the off-site sources AI engines rely on for B2B firms for which ones deserve effort.
The first twelve weeks, week by week.
A well-run engagement spends two weeks measuring, the next few removing blockers, and the rest writing pages and fixing sources, with a re-test every four weeks. Your dates will differ; the order should not.
| Weeks | The agency is doing | You should see |
|---|---|---|
| 1–2 | Gathering real buyer questions from your sales team; running the baseline; requesting read-only analytics access | The question list for sign-off, then the baseline with raw answers |
| 2–4 | Checking crawler access, indexing and rendering; listing wrong facts about your company in answers and on profiles | A blocker list with owners; a fact sheet of corrections for you to approve |
| 3–6 | Rewriting the first pages against specific questions; setting up tracking of AI referral visits | Drafts to approve; a first re-test at week 4 |
| 5–8 | Working through the third-party source list; publishing approved pages | Live pages; profile changes logged; second re-test at week 8 |
| 9–12 | More pages; follow-ups on sources; reading the three test rounds together | A twelve-week review: what moved, what did not, what to stop |
This is the AI-answers layer of an engagement. The wider onboarding any search agency should run is set out in our month-by-month guide to the first 90 days with an SEO agency.
The deliverables to expect.
By the end of the first quarter you should be holding seven documents. If you are paying a retainer and cannot find most of them in your shared folder, ask for them by name.
- The question panel. The agreed list of buyer questions, grouped by stage: problem, shortlist, comparison.
- The baseline and each re-test. Every question, engine, date and country, with the full answer saved.
- The technical access note. What AI crawlers and search engines can and cannot fetch, with fixes.
- The company fact sheet. One approved version of your name, services, locations, founding date and proof points, used everywhere.
- The page plan. Pages to rewrite or create, by URL, each matched to questions from the panel.
- The source list. The third-party sites engines cite for your category, your status on each and the next action.
- The monthly report. What shipped, what the re-test showed and what AI referral visits did in your analytics.
A dashboard login is not on the list. Tools help run questions at volume, and our comparison of AI visibility tracking tools explains how to judge them, but a subscription is an input to the work, not a deliverable.
What the monthly report should show.
A useful report states how often you were named across repeated runs of the same questions, shows the answers behind that number and lists what was shipped. It does not report a “ranking position” in an AI answer, because that number is not stable. We explain how an AI visibility score is calculated in a separate guide.
The evidence for that is a study by SparkToro and Gumshoe.ai, reported by Search Engine Journal on 30 January 2026. It ran 12 brand-recommendation prompts 60 to 100 times each per platform. ChatGPT and Google’s AI each returned the same list of brands in fewer than 1% of repeated runs. The same list in the same order appeared in fewer than 0.1%.
| Outcome across repeated runs | How often it happened | What it means for a report |
|---|---|---|
| Same list of brands | Fewer than 1 run in 100 | One screenshot proves nothing, good or bad |
| Same list in the same order | Fewer than 1 run in 1,000 | “Position 2 in ChatGPT” is not a measure |
| A brand’s share of runs | More consistent than position | Report share of answers over many runs |
So ask for three things in every report: the share of answers that name you, per engine, with the run count; Google’s own Generative AI performance report in Search Console, which its guidance recommends; and visits and enquiries from AI referrals in your analytics. If you have never seen where you stand, the free AI visibility report is a quick way to get a first reading.
When the numbers need to go upward, our guide on presenting AI visibility to a board shows how to fit them on one page.
A realistic timeline: what moves when.
Some results are in your control and arrive in weeks; others depend on engines and third parties and take months, with no promised date. We have not found a timetable from any AI company for when a changed page will appear in its answers, so the stages below are our estimates, not sourced durations.
| Stage | What can change | Who controls it |
|---|---|---|
| First month | Crawler blocks removed; wrong facts corrected on your site and main profiles; baseline in hand | You and the agency. These are checkable on the day they are done |
| Months two and three | Rewritten pages live and indexed; first signs of those pages being cited for specific questions | Shared. You control publishing; engines decide what they cite |
| Months three to six | Share of answers rising across the panel; new reviews and listings appearing; AI referral visits visible in analytics | Mostly engines and third parties. Expect uneven movement between engines |
Being named is a step, not the goal. A report that shows more mentions and no change in enquiries deserves a hard conversation, and our piece on why being cited is not the same as being chosen explains what to look at next.
Real work or relabelled SEO?
Most of GEO is good SEO. The problem is an agency that changes the label and the invoice but adds none of the work specific to AI answers: testing questions in the engines, repeat runs, source research and fact correction. We explain what a credible AI SEO service includes in a separate guide.
Google is direct about the overlap. Its AI features documentation says there are “no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary”, and that a page must be indexed and eligible to appear with a snippet. Its generative AI guide tells readers to prioritise effective SEO strategies over “AEO/GEO hacks”.
| Area | Real GEO or AEO work | Relabelled SEO |
|---|---|---|
| Starting point | Buyer questions tested in the engines, with saved answers | A keyword list with “AI” in the column header |
| Measurement | Share of answers over repeated runs, per engine | Keyword rankings, or one unexplained “AI score” |
| Technical | Crawler access, indexing and rendering checked and fixed | An llms.txt file and “AI schema” sold as the main deliverable |
| Content | Named pages rewritten to answer named questions, with facts you approved | A monthly quota of blog posts on loosely related topics |
| Off-site | A researched list of sources engines cite for your category | A link-building package under a new name |
| Honesty about SEO | Says plainly which tasks are ordinary SEO | Claims everything is new and proprietary |
The last row is the simplest test. An agency that says “much of this is SEO you should already be doing, and here is what is new” understands the work. Google’s hiring advice says the same: “Be careful if a company is secretive or won’t clearly explain what they intend to do.”
Red flags once the work has started.
These six warning signs show up after signing, in months two and three.
- The question list keeps changing. If the questions tested in month three are not the ones in the baseline, nothing can be compared.
- Reports show screenshots of wins only. Answers vary between runs, so ask for every run.
- Volume replaces relevance. Google’s spam policies name “using generative AI tools or other similar tools to generate many pages without adding value for users” as scaled content abuse.
- Reviews or mentions appear that you did not earn. The US Federal Trade Commission’s final rule, announced on 14 August 2024, bans fake reviews and testimonials, including buying them.
- They claim inside knowledge. Google’s guide states: “No third-party tool has access to our internal ranking or AI systems.”
- They ask for more access than the job needs. For an audit, Google advises granting read access to Search Console only.
One flag is a question to ask. Two or more in a month is a reason to request the seven deliverables from chapter five.
Is this agency doing real GEO work? The scorer.
Use this on a proposal or on an agency you already pay. Tick the evidence you have seen in writing, then any warning signs. You get a score out of 100 and a list to send back. Nothing you tick is sent anywhere.
Score: 0 / 100
Tick what you have seen in writing. Anything promised in a call but not written down counts as missing.
The weights are our own judgement, not measured data. They favour the baseline, the page plan and the source list: the hardest to fake and the first to be skipped.
Six questions before you sign.
These six are specific to AI search work. Google’s own advice on hiring adds a seventh worth borrowing: ask whether the agency’s advice on AI experiences is “aligned with Google Search’s official guidance”.
- Which of our buyers’ questions will you test, and can we approve the list? Good answer: a draft list for you to edit.
- How many times do you run each question, in which engines, from which country? Good answer: numbers, not “we use a platform”.
- What will we receive in the first thirty days? Good answer: a baseline and a blocker list, both yours to keep.
- Which parts of this scope are ordinary SEO? Good answer: a straight split, with reasons.
- How will you get third-party mentions, and what will you never do? Good answer: named methods, and a clear no to paid or invented reviews.
- How will we know it produced enquiries? Good answer: AI referral tracking in your analytics.
For the wider interview, including how to check a case study, use our list of 20 questions to ask an SEO or AI search agency. Price depends on scope, so it is not covered here; what drives it is in our guide to what AI search optimisation costs.
If you would like this work scoped against your own buyers’ questions, our answer engine optimisation (AEO) programme starts with the same dated baseline.
Questions buyers ask.
What does an AEO agency do?
An AEO agency works to get your company named and cited in AI-generated answers. It tests the questions your buyers ask in engines such as ChatGPT, Perplexity and Google’s AI features, fixes crawler access, rewrites pages to answer those questions, improves third-party sources and reports the change against a dated baseline.
What is the difference between a GEO agency and an AEO agency?
There is no practical difference. GEO (generative engine optimisation) and AEO (answer engine optimisation) describe the same service, and many agencies use both labels. Google’s guidance treats both as part of SEO. Judge an agency by its deliverables, not by which acronym it uses.
How long does it take to see results from a GEO agency?
Fixes you control, such as crawler access and corrected facts, can be done in the first month. Being cited for specific questions tends to follow page changes over the next two to three months. We have found no published timetable from any AI company, so treat any date as an estimate.
Do we need a separate AEO agency if we already have an SEO agency?
Not necessarily. Much of the work is SEO your current agency should already do. Ask them for the AI-specific parts: a baseline of buyer questions, repeat runs, a source list and AI referral tracking. If they deliver those, a second agency adds little.
Can an AEO agency promise we will appear in ChatGPT answers?
No. Nobody outside the AI companies controls what an engine says, and answers change between runs. A study reported by Search Engine Journal in January 2026 found ChatGPT returned the same brand list in fewer than 1% of repeated runs. An honest agency commits to work and measurement, not placement.
Sources: Prolinkage’s own study of sources cited in 36 AI answers, November 2026. External: Google Search Central, optimising for generative AI search; Google Search Central, do you need an SEO?; Google Search Central, AI features and your website; Google Search Central, spam policies; OpenAI, overview of OpenAI crawlers; Aggarwal et al., “GEO: Generative Engine Optimization”, arXiv, 16 Nov 2023; G2, “The answer economy”, 15 Apr 2026; Search Engine Journal on SparkToro and Gumshoe.ai, 30 Jan 2026; US FTC, final rule banning fake reviews and testimonials, 14 Aug 2024. All accessed 12 November 2026.