Answer Engine Optimization: What Works and How to Prove It

Answer engine optimization (AEO) is making your pages eligible to be retrieved, trusted and named in answers from AI systems such as ChatGPT, Perplexity, Gemini, and Google's AI Overviews and AI Mode. The work overlaps search engine optimization (SEO) almost completely. The difference is the outcome you measure.

AI Overviews are the AI summaries Google shows above results. AI Mode is Google's conversational search experience. Generative engine optimization (GEO) is another name for the same work. Searches that pair the term with "voice search" refer to an older meaning, optimizing for spoken answers from voice assistants, which this page does not cover.

Two outcomes matter. An answer names your brand when it appears in the answer text. An answer cites you when it links your page as a source. We use "named" and "cited" in exactly these senses from here on.

Google and OpenAI document access and quality, not tricks. A page must be crawlable and indexed, and original content helps. Google says no special file, markup or writing style is needed, and OpenAI separates the crawler that governs ChatGPT search from the one used for training. The table checks common tactics against Google's AI optimization guide, Google's AI features documentation and OpenAI's crawler documentation, as of October 2026.

What remains once the folklore is gone is plain: be reachable, say something others cannot, and measure the result.

Yes, if your robots.txt allows OAI-SearchBot and Googlebot for the section you want cited. Run a check for eight crawlers; you pass when both show ALLOW and fail when either shows BLOCK. A robots.txt file sits at yourdomain.com/robots.txt and tells crawlers, the programs that fetch pages, which paths they may fetch.

import urllib.request, urllib.robotparser, urllib.error
BOTS = ["OAI-SearchBot", "GPTBot", "ChatGPT-User", "PerplexityBot",
        "ClaudeBot", "Claude-SearchBot", "Google-Extended", "Googlebot"]
SITES = ["yourdomain.com"]   # replace with your domain
PATH = "/blog/"              # a section you want cited
for site in SITES:
    req = urllib.request.Request(f"https://{site}/robots.txt",
                                 headers={"User-Agent": "Mozilla/5.0"})
    try:
        text = urllib.request.urlopen(req, timeout=15).read().decode("utf8", "ignore")
    except urllib.error.HTTPError as e:
        print(site, "robots.txt unreachable, status", e.code); continue
    except urllib.error.URLError as e:
        print(site, "robots.txt unreachable:", e.reason); continue
    rp = urllib.robotparser.RobotFileParser()
    rp.parse(text.splitlines())
    for bot in BOTS:
        ok = rp.can_fetch(bot, f"https://{site}{PATH}")
        print(site, bot, "ALLOW" if ok else "BLOCK")

ClaudeBot is Anthropic's training crawler and Claude-SearchBot is its search crawler, per Anthropic's crawler documentation. Python's parser ignores wildcard rules, so for Googlebot also check the robots.txt report in Search Console (Google's free site-reporting tool).

We ran this check on October 14, 2026, against a sample of 21 AEO and SEO publishers and tool vendors: airops.com, frase.io, llmrefs.com, greenflagdigital.com, position.digital, cxl.com, digitalapplied.com, almcorp.com, prometheusagency.co, technologyadvice.com, seorecheck.com, ranknext.io, searchengineinfo.org, answerenginesoptimization.ai, nogood.io, shopify.com, semrush.com, ahrefs.com, searchengineland.com, moz.com and hubspot.com. Twenty returned a robots.txt file; searchengineland.com answered with a 403 error, which the script reports as unreachable. In those 20, all eight crawlers were allowed on /blog/ except GPTBot at moz.com. Our own pagelens.ai allows all eight.

Nearly every site in the sample allows these crawlers, so passing the test only means you are eligible. The sample had no group of sites that AI engines ignore, so it cannot tell cited sites from uncited ones. It reads robots.txt only; a firewall or content delivery network can still block a crawler that robots.txt allows, so also look for OAI-SearchBot requests in your server logs.

The same engine gives different answers to the same question, and different engines disagree with each other. In Citation Research Study 001, our research lab's test of 800 answers, only 20% of 400 repeat pairs named exactly the same providers.

The study asked 100 questions about 20 Indian exams on ChatGPT (search on), Perplexity, Gemini and Google AI Mode, twice each in a new session, with fieldwork on October 5, 2026. The first-named provider matched in 64% of the 374 repeat pairs that named providers, and all four engines shared a top pick for only 5 of 20 exams.

Engines also lean on different kinds of source. Study 007 asked 150 decision questions from India, which gave 150 answers per engine, and counted those citing a community platform (Reddit, YouTube or Quora): 117 of 150 from Google AI Mode, 73 from Perplexity, 29 from Gemini and 3 from ChatGPT.

Both studies used English questions phrased the way people in India ask them, on a single day, so they show how variable the engines are, not US rates. Our inference: treat each engine as its own channel, repeat every prompt, and track the first-named brand separately, since it was the steadier signal.

Measure AEO as an appearance rate: the share of collected AI answers that name your brand, calculated per engine from repeated runs of a fixed prompt list. Collect at least 30 answers per engine before reading a rate, and trust only changes larger than the margin for comparing two rounds. These steps are our recommendation.

Four engines at 20 prompts and 3 runs is 240 answers per round, which is the real cost of doing this by hand.

We run the sample on a schedule: each tracked prompt goes to each engine at a fixed cadence, and we record whether your brand is named and which sources are cited. We report share of voice (the share of tracked answers that name you rather than competitors), citation sources and sentiment (how favorably answers describe your brand) against competitors. Plans differ in cadence and engines. Prices are monthly in US dollars as of October 2026 on our pricing page.

• Launch, $299. 100 tracked prompts weekly, 300 AI answers per week, on ChatGPT, Google AI Mode and Perplexity. It includes Prompt Research for choosing prompts, competitor analysis, 25 content pieces per month and unlimited users.
• Growth, $699. 100 prompts daily, 500 answers per day, adding Gemini and Grok, plus technical recommendations and 50 content pieces per month.
• Enterprise, $1,499. 200 prompts daily, 1,400 answers per day, on ChatGPT, Claude, Gemini, Perplexity, Grok, Copilot and Google AI Mode, with the Outreach Agent, technical audit and fixes, and a dedicated account manager.
• Agency, custom pricing. For teams managing several client brands, with a shared prompt pool and white-label hosted publishing.

Our arithmetic from those limits: each plan sends one answer per prompt per engine per cycle (100 × 3 = 300 per week on Launch; 100 × 5 = 500 per day on Growth; 200 × 7 = 1,400 per day on Enterprise). Pooled across 100 prompts, Launch yields 100 answers per engine per week: a single-week margin of about 9 points at a 30% rate (1.96 × √(0.3 × 0.7 ÷ 100) = 9.0), and about 13 points for comparing two weeks. Growth yields 700 per engine per week, which narrows the week-to-week comparison to about 5 points (1.96 × √(2 × 0.3 × 0.7 ÷ 700) = 4.8). Both treat answers as independent, so they are minimums.

They show up in two places, and neither shows the prompt. Search Console's Generative AI performance report covers AI Overviews and AI Mode and reports impressions by page, country, device and date. It has no click or query data, and it rolled out worldwide as of August 31, 2026. Clicks still count inside your ordinary Web search totals, per Google's AI features documentation.

Clicks from chat engines arrive in GA4 (Google Analytics 4) as referrals, and some arrive with no referrer and look like direct traffic. Set up AI traffic channels in GA4 so the visible part is separated. Then read appearance rate, impressions and sessions together: a rising appearance rate with flat impressions and sessions means checking whether answers cite you or only name you.

Run the robots.txt test, confirm indexing in Search Console, and publish first-hand material only you can write. Then take four weekly baselines with the numbered steps above before you change anything.

If you would rather not run 240 answers by hand every week, Book a demo to see how PageLens.ai runs the sample on a schedule.