GEO vs SEO: What Actually Changes, Plus a 60-Minute Test

GEO is mostly the same work as SEO. Google's documentation states: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." A page needs to be indexed and eligible to appear in Google Search with a snippet. Three parts of the job do change, and they are covered below.

Google's AI features documentation says existing SEO fundamentals remain worthwhile: pages crawlers may fetch, good internal linking, page experience, text content, accurate structured data (markup that labels a page's facts for machines) and current business information. It also says you do not need new machine-readable files, "AI text files," or special schema.

On May 15, 2026, Google went further with a dedicated guide to optimizing for generative AI features. It points to the same priorities: unique, first-hand, people-first content and sound technical basics. It says you do not need llms.txt (a proposed plain-text file that lists a site's key pages for AI assistants), special markup, Markdown versions, or content cut into tiny chunks. It adds that structured data is not required for generative AI search.

So the objection that "it's all search" is right about Google's own AI features. Anyone selling a special file or markup as the fix for them is selling something Google says is not required.

The objection is incomplete elsewhere. ChatGPT and Perplexity run their own retrieval, and Google's guidance does not describe how they behave. Do not assume rankings carry over to them. Test.

Three things change: whether AI crawlers can reach your site, whether your content covers the sub-questions behind a search, and how you measure success. Indexing, useful content and links carry over unchanged. The table shows each area, what to do and the check that tells you whether it works.

Less often than a top-10 ranking would predict. In a March 2026 study by Ahrefs, which sells SEO and AI-visibility software, 37.9% of cited URLs ranked in the top 10, 31.2% ranked 11 to 100 and 31.0% ranked beyond 100. The sample was 863,000 keywords and 4 million cited URLs; the shares sum to 100.1% because of rounding.

An earlier Ahrefs study from July 2025 looked only at the three most visible citations per AI Overview (1.9 million citations from 1 million AI Overviews) and found 76.1% ranked in the top 10. Because the earlier study counted three citations per answer and the later one counted every cited URL, the two are not directly comparable. The direction is a signal; the size of the drop is not.

Ahrefs links the shift to query fan-out: Google splits a search into several related sub-queries, and the pages that show up most often across those results get cited. A page can be cited without ranking for the exact search. For you, that means writing for the cluster of questions around a topic. Our guide to ranking in Google AI Overviews covers the page-level steps.

The 2024 research paper that introduced the term, GEO: Generative Engine Optimization, tested content changes on a benchmark and reported visibility gains of up to 40%, varying by domain. That is a ceiling measured on a research benchmark, not a forecast for your site.

These figures cover Google AI Overviews or a research benchmark. For ChatGPT and Perplexity, assume nothing about rank and read what they actually cite.

Measure your own numbers, because there is no reliable industry figure for AI search traffic. What is measurable is behavior on Google: Pew found a regular result was clicked in 8% of visits when an AI summary appeared and 15% when none did. The study tracked 900 US adults in March 2025 across 68,879 searches, 18% of which showed a summary.

Users clicked a link inside the summary in 1% of visits and ended their session in 26% of visits with a summary against 16% without. A brand named in an answer that nobody clicks can still shape the shortlist, so mention rate belongs next to traffic. Our guide to AI search analytics lists the metrics worth tracking.

As of October 2026, Google's Search Console Search Generative AI performance reports show impressions for your pages in AI Overviews and AI Mode by page, country and device, with data from May 18, 2026. They do not show clicks, click-through rate or queries, and they are rolling out gradually, so your property may not have them yet. They cover Google only, so they do not replace the prompt log below for ChatGPT, Perplexity and other assistants.

Avoid four: treating a ranking as a citation forecast, judging from one answer, counting only clicks, and editing only your own pages. Our recommendation is to check the layers in order, because a crawler that cannot reach a page, or a page that is not indexed, cannot be cited.

Run checks one and two first and fix any failure before continuing, because AI engines cannot cite pages they cannot reach or that Google has not indexed. Then run checks three to five even if you fail them, because they show why. For a fuller version, see our guide to running an AI visibility audit.

Our time model for checks one to four: script on three domains, 5 minutes; Search Console on 5 key URLs, 5 minutes; 30 prompt runs at about a minute each, including logging, 30 minutes; labeling cited URLs and checking rankings, 20 minutes. The total is about 60 minutes (our estimate). The repeat in check five takes about 30 minutes, a week later.

Can AI crawlers reach your site?

Run this script on your domain and on two competitors. Pass: no root-level Disallow: / against a bot you want to reach you. Fail: a block you did not choose. Blocking can be deliberate, so check each company's crawler documentation for what each bot does before you allow or block it. robots.txt is a request that crawlers are expected to honor, not a lock.

# aibots.py   usage: python3 aibots.py yourdomain.com competitor.com
import sys, urllib.request
BOTS = {"gptbot","oai-searchbot","chatgpt-user","claudebot","claude-searchbot",
        "claude-user","perplexitybot","perplexity-user","google-extended",
        "ccbot","anthropic-ai"}
def blocked(domain):
    req = urllib.request.Request(f"https://{domain}/robots.txt",
                                 headers={"User-Agent": "Mozilla/5.0"})
    text = urllib.request.urlopen(req, timeout=15).read().decode("utf-8", "ignore")
    groups, agents, rules, last_was_agent = [], [], [], False
    for line in text.splitlines():
        line = line.split("#")[0].strip()
        if ":" not in line: continue
        key, val = [s.strip() for s in line.split(":", 1)]
        key = key.lower()
        if key == "user-agent":
            if not last_was_agent and agents:
                groups.append((agents, rules)); agents, rules = [], []
            agents.append(val.lower()); last_was_agent = True
        else:
            last_was_agent = False
            if key in ("allow", "disallow"): rules.append((key, val))
    if agents: groups.append((agents, rules))
    hit = set()
    for agents, rules in groups:
        if ("disallow", "/") in rules:
            hit |= {a for a in agents if a in BOTS}
    return sorted(hit)
for d in sys.argv[1:]:
    try: print(d, blocked(d) or "no root-level block")
    except Exception as e: print(d, "could not fetch:", e)

The script flags only the eleven named agents. A group that also lists Allow lines for some paths still counts, so read it by hand. A User-agent: * group with Disallow: / blocks every crawler and also needs a manual read.

One distinction matters. According to Google's crawler documentation, Google-Extended controls whether your content may be used for training future Gemini models and for grounding in Gemini Apps and Vertex AI. It does not affect inclusion in Google Search or act as a ranking signal. Do not read a Google-Extended block as a reason you are missing from AI Overviews or AI Mode.

On October 9, 2026, we ran the script on the robots.txt files of 12 sites: pagelens.ai, nytimes.com, cnn.com, theverge.com, builtin.com, forbes.com, hubspot.com, ahrefs.com, semrush.com, moz.com, writesonic.com and wordstream.com. Four returned a root-level Disallow: / for at least one of the eleven agents. nytimes.com and cnn.com did so for all eleven, theverge.com for nine (while allowing /sp/ paths) and builtin.com for four. Eight returned none. This is a convenience sample of news and marketing sites, not a market measure.

Are your key pages indexed and eligible for a snippet in Google?

Paste each key URL into Search Console's URL Inspection tool. Pass: the page is on Google and carries no noindex or nosnippet directive. Fail: anything else. Google's condition for appearing in AI Overviews and AI Mode is indexing plus snippet eligibility, so a failure here caps everything below.

Which buyer prompts name you?

Write 10 prompts a buyer would type, such as "best [category] for [use case]." Ask each in ChatGPT, Perplexity and Google AI Mode. That is 30 runs. For a quick first read without building a log, our free AI visibility audit on our homepage takes a domain and returns results in about 30 seconds. Otherwise, log one row per run:

Mention rate = runs naming your brand ÷ total runs. Citation rate = runs citing a URL on your domain ÷ total runs. Both are our working definitions, not an industry standard. Pass: your mention rate is at least that of the most-named competitor across the same prompts. Fail: it is lower.

Do the cited pages explain your gap?

For every prompt where you were not named, label each cited URL: your domain, a competitor, a third-party list or review, a video or forum, or other. Then pick the closest short keyword for the prompt (for example, "best [category]") and check whether you rank in the top 10 for it in Search Console or a rank tracker. Pass: no prompt where you rank top 10 for its keyword yet are never cited. Fail: any such prompt, which is a coverage gap. Add a section or page for the related questions the cited pages answer and yours does not.

Does the result hold when you ask again?

Repeat the same 30 runs a week later. For each prompt-and-engine pair, you are named in both rounds (stable), in one (unstable) or in neither. Stability = stable pairs ÷ pairs where you were named at least once. Our rule of thumb: below about two-thirds, treat your visibility as unstable. Pass: stability at or above that. Fail: below it. Our guide to tracking AI citations over time covers how to keep this up.

Reading the results. Our recommendation:

• Blocked or not indexed: fix access first.
• Indexed and open, ranking but uncited: close the coverage gap.
• Neither ranking nor cited: do standard SEO work first.
• Named once but not on repeat: treat it as unstable and keep tracking.

We run the same prompt log automatically. You choose the buyer prompts, and we ask them across AI engines on a schedule and record who is named, which URLs are cited, sentiment (the tone and adjectives AI answers use about your brand) and share of voice (your brand's share of all brand mentions across the same prompts), compared with your competitors. We also report human and AI crawler traffic to your site. Our log records what the engines answer; it does not read Google's internal data.

When the log shows a gap, our content agent writes the page that answers it and publishes it on your own domain. As of October 2026, our US plans are billed monthly in dollars and every plan includes unlimited users:

• Launch, $299 a month: 100 tracked prompts checked weekly on ChatGPT, Google AI Mode and Perplexity, with 25 content pieces a month.
• Growth, $699 a month: 100 tracked prompts checked daily on five engines (adds Gemini and Grok), 50 content pieces a month, and technical recommendations.
• Enterprise, $1,499 a month: 200 tracked prompts checked daily on seven engines (adds Claude and Copilot), 100 content pieces a month, a dedicated account manager, and an Outreach Agent that drafts pitches to sites already cited by AI answers, which you approve before anything is sent.

Agency pricing is custom. Details are on our pricing page. To see the log run on your own prompts, Book a demo.