Key Takeaways
- Five minimum technical conditions determine whether you should hire AEO agency in Vietnam. You can check all five yourself in about 10 minutes.
- Six variables drive cost when hiring an AEO agency. The one covering external entity signals is the most overlooked and has the largest effect.
- 85.7% of AI citations about a brand point to third-party sources, only 14.3% to the brand’s own site, per research across 167,551 citations with open data.
- Any AEO agency guaranteeing a specific citation rate is a warning sign, because AI output is probabilistic.
- Technical changes show results in 4 to 12 weeks, entity signals take 3 to 6 months. Measure monthly, conclude quarterly.
What Your Website Needs Before Hiring an AEO Agency
AEO is an optimisation layer sitting on a technical foundation. If the foundation fails, no amount of content quality reaches the AI crawler.
Check this before discussing budget to hire an AEO service or hire an AEO agency, because it determines what service you are actually buying. You can verify all five conditions below yourself without deep technical knowledge.
1. Content is server-side rendered
How to check: disable JavaScript in your browser (Chrome: Settings, Privacy and security, Site settings, JavaScript, Don’t allow) and reload the page. If the content disappears or leaves an empty shell, your site is client-side rendered.
Why it matters: many AI crawlers do not execute JavaScript. To those crawlers, your page is empty. This is a prerequisite, not a bonus.
2. No lingering noindex tags
How to check: view page source and search for noindex. Also check yourdomain.com/robots.txt.
Why it matters: a page blocked from indexing never enters Google’s index, which means it can never appear in AI Overview. This is more common than people expect, usually left over from the development phase.
3. Ability to inject JSON-LD into the head
How to check: ask your technical team, or try adding a schema block to one page. If your CMS does not allow per-page intervention in <head>, that is a constraint to resolve.
Why it matters: schema must be delivered server-side. Some platforms only allow scripts in the body, or inject after page load, both of which make schema invisible to crawlers that skip JavaScript.
4. Heading structure that reflects actual content
How to check: open a page and confirm there is exactly one H1, and that H2s are genuine section headings rather than styling for large text.
Why it matters: AI decomposes a page along its heading structure to find the passage answering each sub-question. Headings used for styling break that decomposition.
5. Stable load speed
How to check: PageSpeed Insights, look at Core Web Vitals.
Why it matters: crawlers have timeout limits. Pages that load too slowly can be skipped on some crawl passes.
If your website does not meet these conditions
This is not a content problem, it is a web development problem. And this is why Brandgineer combines SEO/AEO with Web Development into one system rather than splitting them into separate services: in practice most AI visibility problems originate in the technical layer.
When those two parts sit with two different vendors, the common outcome is that the content side says the website is blocking results, the technical side says the content is not good enough, and the business in the middle cannot tell where the real problem is. That is a hidden cost rarely factored into quote comparisons.
This does not mean you must choose an integrated firm. If your website already meets all five conditions, a specialist AEO agency is sufficient and may cost less. The point of checking these five conditions is to know which situation you are in before you negotiate.
Six Variables That Drive AEO Service Cost When Hiring an AEO Agency
This article does not quote figures, because pricing of AEO agency depends heavily on scope. Instead, here are the six variables that move it, so you can estimate for yourself and compare quotes on the same basis when hire an AEO agency.
1. Number of prompts tracked
The most basic variable of every AEO service. Tracking 15 prompts is a different workload from tracking 100.
When comparing quotes, ask how many times each prompt is run per measurement cycle. A firm quoting 50 prompts at one run each and a firm quoting 20 prompts at five runs each are selling two different things, and the second gives more reliable data.
2. Number of engines measured
ChatGPT, Google AI Overview, Gemini, Perplexity. Each engine is a separate run and a separate dataset, because they ground in very different source sets.
3. Number of languages
If your customers only ask in Vietnamese, one language is enough. If you have international customers or want to cover ChatGPT properly, you need an English prompt set too, and measurement cost roughly doubles.
4. Ratio of new content to optimising existing content
Reworking an existing article is far cheaper than researching and writing new. If you already have a content library, most of the early value comes from improving structure rather than producing more.
This is worth probing: a proposal that is entirely new articles when you already have 50 unoptimised ones suggests the current state was not surveyed.
5. Depth of technical intervention
From adding schema to an existing template, to moving to server-side rendering, to rebuilding the site. These three levels differ enormously in cost and time.
6. Building entity signals outside your website
The most overlooked variable, and the most important one.
Research across 167,551 URL-grounded citations covering 128 brands in 12 markets found 85.7% of citations about a brand point to third-party sources, with only 14.3% pointing to the brand’s own site. McKinsey research reports an equivalent ratio. Bain says the same thing differently: LLMs reach for reviews, earned media and comparison sites to describe a brand.
Which means a quote covering only content on your own website is targeting roughly 14% of what AI reads.
Entity signal work covers PR and industry publication placements, completing directory and comparison-site profiles, and building presence on the platforms AI actually reads in your sector. It costs more and moves slower than content, but skipping it puts a low ceiling on results.
A note specific to the Vietnamese market: the same research found that in languages with a thinner native web, the share of AI citations coming from the brand’s own site is notably higher than in languages with a dense web. Vietnamese sits in the thin-web group, so the 14% figure is likely higher in practice in Vietnam. This is an inference from comparable-language data rather than a direct measurement on Vietnamese, so treat it as a hypothesis to verify with your own industry data.
What a Real AEO Audit Contains
An audit is what you receive before hiring an AEO agency, which makes it the best available sample of that firm’s actual work quality.
Four required components of an AEO audit
1. Baseline AI visibility per engine, with method stated
Not just a number. It must include: which prompt set, in which language, on which engines, how many runs per prompt, and when it was measured.
Missing the “how many runs” part is a serious omission. AI output is probabilistic; one run is not one result. Professional measurement platforms sample each prompt dozens to hundreds of times for exactly this reason.
2. Technical state and machine readability
Existing schema markup, server-side rendering test results, index status. This section should match the five conditions at the top of this guide, and you can re-verify it yourself.
3. Which sources AI currently cites for your industry queries
This is the most valuable component and the most commonly missing one. Not “where you stand” but “who AI is reading to answer questions about your industry”. That list is the map of where you need to be present.
4. The gap against competitors on the same prompt set
Your own number means nothing without context. A 5% citation rate could be good or terrible depending on where competitors sit.
Signs of a going-through-the-motions audit
- Only generic technical errors that any automated tool would output
- No specific prompt set stated, just “we analysed the market”
- A single composite score with no explanation of how it was calculated
- Nothing about what you are doing well, only problems to fix
- No statement of how many runs per prompt
The last point is the most telling, because it separates firms that understand AI’s probabilistic nature from firms running a tool and exporting a report.
What Is Reasonable for AEO KPI Commitment
A note worth considering if you want to hire an AEO agency: No agency controls an AI engine’s algorithm. So reasonable commitments cover work and method, not a result figure.
Reasonable commitments before you hire an AEO agency
- Volume of work delivered: articles optimised, schema implemented, new articles, entity signal activities
- A fixed reporting cadence with report contents defined in advance
- A fixed measurement method that does not change midway: same prompt set, same run count, same engines
- Response time for technical issues
Commitments that warrant caution
- A specific citation rate or visibility score within a specific timeframe
- A guarantee of “reaching AI Overview” for a specific query
- A mention-count guarantee with no measurement method attached
The reason is simple: an agency guaranteeing a 10% citation rate is promising something dependent on Google’s and OpenAI’s algorithms, on competitor behaviour, and on the inherent randomness of the models. Nobody controls all three.
How to test it in the meeting: ask directly, “if the committed figure is not met after six months, what happens?” The answer tells you whether the commitment is real or sales language.
KPIs worth tracking when you hire an AEO agency
Three core metrics: AI citation rate, AI visibility score, AI share of voice. Plus two supporting metrics that matter more in the early phase: the count of new prompts where you start appearing, and AI-referred sessions in GA4.
The detail on measuring each is covered separately: How to measure your brand in AI search.
Realistic Timelines
Three different timeframes for three different kinds of work. Collapsing them into one is the source of most mismatched expectations.
Technical and structural changes: 4 to 12 weeks. Schema markup, content structure improvements, fixing server-side rendering. This produces results soonest because it directly affects whether content can be retrieved at all.
New content: 8 to 16 weeks. New articles need time to be crawled, indexed and to accumulate linking signals. In real deployments, homepages and service pages tend to get cited before new blog posts, even when the blog posts are better optimised.
Entity signals: 3 to 6 months. The slowest part and the most important. AI model perceptions of a brand shift on roughly a 3 to 6 month cycle.
What this means for evaluation: measure monthly to catch problems early, but draw conclusions about programme effectiveness quarterly. Evaluating an AEO programme after 30 days and concluding it does not work is a methodological error, not a vendor failure.
One notable exception: if 90 days pass with no new prompts where you start appearing, that is no longer a timing issue but a content or entity signal issue. That is a fair point at which to ask serious questions.
Seven Questions to Ask Before hiring an AEO agency
- How do you measure AI visibility, and how many runs per prompt per cycle?
- Which engines do you track, in which languages, and why that set?
- Is the contractual commitment about volume of work or about a result figure?
- If my website does not meet technical standards, how is that handled and who owns it?
- What do you do to build entity signals outside my website?
- What does the monthly report contain, and how long before we draw conclusions?
- In what situation would you advise me not to hire an AEO service
- yet?
Questions 1 and 5 separate real capability most clearly. Question 1 because it touches AI’s probabilistic nature, which many firms have not internalised. Question 5 because it touches the 85% of sources AI actually reads, the hardest part to deliver and the most frequently absent from quotes.
Question 7 tests whether you are getting advice or a sale. A firm that would never tell a client to hold off is selling, not advising.
FAQ
What does my website need before AEO can work?
Five minimum conditions: content is server-side rendered, no lingering noindex tags, the ability to inject JSON-LD into the head, heading structure that reflects actual content, and stable load speed. If any are missing, that is a web development problem to resolve before or alongside AEO.
What drives the cost of an AEO service?
Six variables: number of prompts tracked, number of engines, number of languages, the ratio of new content to optimisation of existing content, depth of technical intervention, and whether external entity signal work is included. The last has the largest effect, because 85.7% of AI citations come from third-party sources.
What should a real AEO audit contain?
Four things: a per-engine baseline with the measurement method stated, technical state and machine readability, the list of sources AI currently cites for your industry queries, and the gap against competitors on the same prompt set. An audit that does not state runs per prompt is a warning sign.
What KPI commitment is reasonable from an AEO agency?
Commitments about volume of work, reporting cadence, and a fixed measurement method. Not a specific citation rate, because AI output is probabilistic and nobody controls the engines’ algorithms.
How long until AEO shows results?
Technical and structural changes: 4 to 12 weeks. New content: 8 to 16 weeks. Entity signals: 3 to 6 months. Measure monthly, conclude quarterly. If 90 days produce no new prompts at all, the issue is content or entity signals rather than timing.
Should I do AEO in-house or hire an agency?
Baseline measurement and content structure improvements are doable in-house with a dedicated content owner. The harder parts to do internally are technical website intervention and external entity signal building, since both require skills and relationships different from day-to-day content work.
