Answer Engine Optimization (AEO)

Answer Engine Optimization (AEO) is the practice of making your content easy for AI systems, such as ChatGPT, Perplexity, Gemini, Claude and Google's AI features, to find, read and cite when they answer questions. Classic SEO works for a ranked list of links. AEO works for the answer itself: the brands an assistant names and the sources it quotes or links.

Hydrogen SEO handles both halves. The free plugin produces clean, server-rendered output that machines can read. The hydrogenseo.com platform checks whether answer engines can use a page, and measures whether they actually name and cite you.

How AEO differs from classic SEO

The two overlap heavily: indexable pages, clear writing and accurate markup help both. The emphasis shifts:

The practical consequence: besides "do I rank?", you also ask "when an assistant answers this question, am I in the answer?" That question needs its own tools.

The three AEO tools on the platform

AEO analysis

Every plan, including Free. At /seo/aeo-analysis, enter a URL and get a score out of 100 with a list of checks. Each check shows what backs it: the engines' own documentation, published studies, or plain good practice. The checks cover:

• AI crawler access. robots.txt is read the way crawlers read it, for 13 AI crawlers, in four groups:
  • Search crawlers (OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot, Bingbot) decide whether a page can appear in answers. Claude's web search also draws on a third-party index (Brave Search), which follows Googlebot's rules.
  • Google-Extended controls whether Gemini apps and Gemini grounding with Google Search may use the page. Blocking it keeps you out of those answers, but not out of Google Search or AI Overviews.
  • User fetchers (ChatGPT-User, Perplexity-User, Claude-User) open a page when someone asks an assistant about it.
  • Training crawlers (GPTBot, ClaudeBot, CCBot, Applebot-Extended) only affect model training, so blocking them is reported as information, not as a problem.
• Firewall test. The page is also requested with the search crawlers' user agents. CDNs and firewalls often refuse AI crawlers by default, and robots.txt cannot show that. The test requests do not come from the crawlers' own IP addresses, so a firewall that verifies bots by IP can still treat the real crawlers differently.
• Indexing and snippets. Google only uses pages in AI Overviews and AI Mode when they are indexed and allowed to show a snippet, so noindex, nosnippet and max-snippet are checked, along with the canonical URL.
• Readable without JavaScript. The text in the raw HTML is compared with the rendered page. OpenAI's, Anthropic's and Perplexity's crawlers do not run JavaScript.
• Answer structure, facts and sources. The checks cover whether the page gets to the point early, whether its sections are focused, and whether it has specific figures, quotations and links to primary sources. In the GEO experiment (KDD 2024), adding statistics, quotations and citations raised a source's prominence in AI answers. The structure checks rest on correlations: patterns in pages that get cited, which cannot prove cause.
• Freshness and authorship. Whether engines can read a publish or update date, and whether an author is named.

Every check is labelled with its evidence, and that sets its weight:

• the engines' own documentation and experiments count in full;
• correlations in studies of cited pages count three quarters;
• plain good practice counts half.

Pages the engines cannot use are capped whatever their content: a page blocked from crawling or indexing scores at most 20, and a page Google may not quote (nosnippet) at most 40.

Some checks are shown but never scored, because the evidence for them is missing or mixed: llms.txt, structured data and question-style headings. See what we do not score below.

An analysis costs 3 credits.

AI visibility

Pro and Max plans. At /seo/ai-visibility, you enter your brand, its website and up to five competitors, and write the questions your customers ask, one per line. Hydrogen SEO sends every prompt to each engine you pick: ChatGPT, Perplexity, Gemini, Claude and Grok. Each engine is asked through its API with the vendor's own live web search turned on, so the answers cite current sources.

You can also click Suggest prompts and competitors (2 credits). It reads your homepage and drafts realistic prompts, most of them without your brand name, plus likely competitors. Check the suggested competitor domains before you run.

Prompts that name your brand ("Is Acme any good?") are reported separately. Answers to them name you almost by default, so they would inflate your visibility. The headline numbers use the other prompts.

The results show:

Beyond the metrics, you get:

• a leaderboard against your competitors;
• a breakdown by engine;
• a grid of prompts by engine, where you can open any answer to read it with the brand names highlighted and its sources listed;
• the most-cited source domains;
• other brands the answers named that you do not track yet.

Brand mentions are matched on the name, the other names you give and the domain, respecting word edges. A domain that only appears inside a citation link counts as a citation, not a mention. A small model reads each answer only for sentiment and for the names of untracked brands.

Each answer costs between 3 and 20 credits depending on the engine (Perplexity 3, ChatGPT and Claude 8, Gemini 10, Grok 20). The form shows the total before you run. Credits for answers that fail are refunded.

AEO monitor

Pro and Max plans. At /seo/aeo-monitor, save a brand, its competitors and a prompt set as a monitor, and it re-runs twice a month (every 15 days) or daily. Each monitor shows:

• the trend for you and each competitor: visibility, citation rate and share of voice;
• every run, with its full answers kept;
• alerts.

Trends only compare runs with the same prompts, brands (names, other names and domains), engines, runs per prompt and country, so changing any of those starts a new trend line.

Alerts fire when:

• your visibility moves, or your citation rate falls, by at least your threshold (15 points by default);
• a competitor pulls ahead of you by at least the threshold;
• two or more answers describe you negatively, after a run where none did;
• a run fails.

Rate changes are measured only on the prompt and engine pairs both runs answered, so an engine that failed in one run does not look like a drop. They must also hold up in a paired sign test, not just pass the threshold. With a small prompt set that means few alerts, and with several checks per run an occasional false alarm is still possible. Alerts appear in the app, and by email if you turn that on.

Free accounts can open both pages and see an example of the results. Running checks and monitors needs an active Pro or Max plan. If a plan lapses or is downgraded, existing monitors pause with an alert, and everything they ran stays readable.

A monitor costs the same per run as the equivalent AI visibility check. The form shows the cost per run and per 30 days. If a scheduled run would need more credits than the account has, the monitor pauses and sends an alert instead of running. You can have up to 10 monitors.

How to read the numbers

• These are API answers. The consumer ChatGPT, Gemini, Claude and Perplexity apps add personalization, memory and their own instructions, so what one person sees there can differ. The API answers are asked the same way every time, which is what makes them comparable over time.
• These are fast models. Each engine uses one of its vendor's fast models, shown in the engine picker, with a few search results per answer. The consumer apps may use larger models and read more sources.
• Engines vary from run to run. The same question can name different brands each time. The headline visibility shows a 95% range, for example 0–24% for 0 of 12 answers. That range covers run-to-run variation on your prompt set only: a different set of prompts can give different numbers, and per-engine or per-prompt numbers rest on fewer answers. Use more prompts, or more runs per prompt under More options, for steadier numbers.
• Google AI Overviews and AI Mode have no public API, and are not tracked. Gemini with Google Search grounding is a different product, and its answers are not a stand-in for them. The AEO analysis checks the page-side requirements for Google's AI features: indexing, snippets and crawler access.
• Country is a hint. It tells the engine where the asker is, which shapes the answer; the web search itself is not localised.

What we do not score

Some popular AEO advice has no measured effect, so it is shown in the analysis but never scored:

• llms.txt is a proposed convention. A study of about 300,000 domains found no link between having one and being cited, and Google says its search does not use it. The plugin's llms.txt editor makes publishing one easy, and it does no harm.
• Structured data is still worth having for rich results; see the schema markup overview. For AI answers the evidence is mixed. A controlled test found no gain in ChatGPT or AI Mode citations after adding it, and a small drop in AI Overviews. Google says its AI features need no special markup, while Microsoft says it helps Copilot understand pages. Google no longer shows FAQ or HowTo rich results at all.
• Question-style headings are often recommended, but in one large correlational study, pages with them were cited about a fifth less often. The same study found question titles helped smaller sites. Use them where they read naturally.

A practical AEO checklist

1. Let the search crawlers in. Allow OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot and Bingbot in robots.txt, and check that your CDN or firewall does not block them. Allow Google-Extended if you want to appear in Gemini answers. To stay out of model training, block GPTBot, ClaudeBot, CCBot and Applebot-Extended instead.
2. Keep pages indexable and snippet-eligible. No stray noindex or nosnippet on pages you want quoted.
3. Put the content in the HTML. Server-render the main text. Hydrogen SEO's own output (titles, meta, canonicals, schema) is always server-rendered.
4. Answer early. Open with a plain paragraph that states the answer or the key facts, then elaborate.
5. Give engines something to quote. Specific figures, named quotations, and links to primary sources.
6. Keep pages fresh and signed. Show an update date and a named author.
7. Measure. Run an AI visibility check on the questions that matter, then set up a monitor to see whether your work moves the numbers.

FAQ

Is AEO a replacement for SEO?

No, it is an extension. Answer engines lean on search indexes and on the same signals of quality and freshness. AEO adds crawler access for AI search bots, quotable content and a new kind of measurement.

Why is my visibility 0% when I rank on Google?

Answer engines name the brands their sources name. Look at Most-cited sources in the results: those are the sites the engines rely on for your prompts. Being covered on those sites is usually what gets a brand into answers.

Do I need the paid platform to do AEO?

No. The free plugin gives you server-rendered output, schema markup and the llms.txt editor with no account. The platform adds the analysis and the measurement. AEO analysis is on every plan, including Free. AI visibility and AEO monitors come with Pro and Max.

Does blocking GPTBot remove me from ChatGPT?

No. GPTBot is OpenAI's training crawler. ChatGPT search uses OAI-SearchBot. The AEO analysis shows both separately.

Is blocking Google-Extended safe?

It does not affect Google Search or AI Overviews. It does keep the page out of Gemini apps and Gemini grounding with Google Search, so it lowers your Gemini numbers in AI visibility.

Related reading

• llms.txt editor
• Schema markup overview
• The hydrogenseo.com platform
• Bulk audit and SEO tools
• What is Answer Engine Optimization?