You asked ChatGPT a question from your own niche. It cited three websites. Yours wasn’t one of them.
GEO (generative engine optimization) is how you fix that. It’s the work of getting your pages cited inside AI answers from ChatGPT, Perplexity, Copilot and Google’s AI Overviews. For most sites, it’s the SEO you already do plus sections a machine can quote on their own.
Paid trackers can wait until a client asks. We ran a GEO check on DroidCrunch, and the weak spot was a file we’d forgotten.
What is GEO, in plain English?
GEO means shaping your content so AI search tools find it, quote it and link to it. SEO wins you a blue link. GEO wins you a numbered citation inside a written answer.
The name comes from a research paper by a Princeton and IIT Delhi team. It went online in November 2023 and was presented at the KDD conference in August 2024. The team built a set of 10,000 queries and measured how small page edits changed a page’s share of an AI answer.
Here’s the part the hype skips. An AI answer engine is still a search engine underneath. It can only cite what it has already found.
So the first job of GEO is old-fashioned: be crawlable, be indexed and rank somewhere for the question. Not sure which questions your pages can win? Our Semrush keyword research tutorial covers that part.
How does an AI answer pick its sources?
An AI answer picks its sources in four steps. It splits your question, searches an index, keeps the passages that fit best, then writes the answer and cites them.
- Split the question. Google calls this query fan-out. Ask about “GEO tools for a small blog” and the system may also search for prices, free options and reviews.
- Search an index. Google’s AI Overviews and AI Mode use Google’s own index. ChatGPT search uses OpenAI’s crawler, OAI-SearchBot, plus outside search providers. Perplexity runs PerplexityBot, and Copilot sits on Bing.
- Pick passages. The model doesn’t read your page top to bottom. It scores chunks, often a few sentences under one heading, against each sub-question.
- Write and cite. The answer is stitched together from those chunks. The citation lands on whichever page supplied the line it needed.
So you aren’t competing page against page. You’re competing passage against passage.
That changes how you write. Bury your key fact in paragraph nine behind “as mentioned above” and you’ll lose to a thinner page. All it has to do is state the same fact cleanly under a matching heading.
What actually moved citations in the GEO study?
Three edits worked best: citing sources, adding quotations and adding statistics. Each lifted a page’s visibility in AI answers by roughly 30-40% on the study’s main measure. On live Perplexity, the best gain was 37%.
Style edits barely registered. Making text more fluent or more authoritative in tone moved results a few points either way. Keyword stuffing didn’t help.
Then comes the twist most GEO articles leave out.
| Edit | Page ranked 1st | Page ranked 5th |
|---|---|---|
| Cite sources | -30.3% | +115.1% |
| Add quotations | -22.9% | +99.7% |
| Add statistics | -20.6% | +97.9% |
The edits that doubled visibility for the fifth-ranked page cut it for the page in first place. Space inside one answer is limited. When lower pages get more quotable, the top page’s share shrinks.
What does that mean for you? If you sit in the lower half of page one, real numbers and named sources are your cheapest lever. If you already rank first, you’re defending, and freshness beats tricks.
Our take: treat the study as direction, not a recipe. It used 2023-era models in a simulated engine, and today’s systems retrieve differently. The core lesson holds anyway: specific, checkable sentences get quoted.
Can the AI bots even read your site?
Check this before you touch a single word. Each AI search tool sends its own crawler, and blocking the wrong one quietly drops you from that tool’s answers.
| Bot | What it does | If you block it |
|---|---|---|
| Googlebot | Feeds Google Search, AI Overviews and AI Mode | You vanish from Google |
| Google-Extended | Controls Gemini training only | AI Overviews still use you |
| OAI-SearchBot | Feeds ChatGPT search | ChatGPT can’t cite you |
| GPTBot | Collects pages for OpenAI training | ChatGPT search still can |
| PerplexityBot | Feeds Perplexity search | Perplexity can’t cite you |
| Claude-SearchBot | Feeds Claude’s search results | Claude’s search skips you |
The classic mistake runs both ways. People block GPTBot to stop training and think they’ve left ChatGPT. They haven’t.
Others block Google-Extended to leave AI Overviews, and stay right in.
Google now has a real switch for that. Since August 31, 2026, every Search Console property has a Search generative AI setting. It includes or excludes your site from AI Overviews and AI Mode.
Robots.txt isn’t the only gate, either. A CDN firewall can return 403 to a crawler your robots.txt welcomes. So we fetched one of our own articles as five different visitors:
$ curl -s https://droidcrunch.com/robots.txt
User-agent: *
Disallow: /wp-admin/
Allow: /wp-admin/admin-ajax.php
# Same article, five user agents, 21 October 2026
Browser 200 399 KB FAQPage schema: yes
OAI-SearchBot 200 399 KB FAQPage schema: yes
GPTBot 200 399 KB FAQPage schema: yes
PerplexityBot 200 399 KB FAQPage schema: yes
ClaudeBot 200 399 KB FAQPage schema: yesEvery bot got the full page, schema included. The article text sits in the HTML, not behind JavaScript, which matters because most AI crawlers don’t run scripts.
One caveat: we faked those user agents from our own server. Cloudflare checks real crawlers by IP address, so your CDN’s crawler logs have the final word.
Does llms.txt help, or is it a placebo?
Mostly a placebo, for now. Google says its AI features need no AI text files or special markup. OpenAI, Anthropic and Perplexity don’t say their crawlers read llms.txt either.
We have one anyway, because Rank Math builds it with one toggle. Reading ours line by line was the most useful ten minutes of this audit.
It lists only our 50 newest posts, out of 577 published. Worse, 32 of those 50 entries use the post’s opening hook instead of a summary.
Our Semrush pricing post shows up as “If you last looked at Semrush pricing in 2025, your numbers are out of date.” Great first line for a person. As a summary for a machine, it says nothing.
Why? Rank Math builds each line from the post’s excerpt, and with no excerpt it grabs the first sentence. Our 18 posts with a written excerpt got proper summaries.
So keep the file if your plugin makes it free, and write excerpts. Just don’t expect it to move citations by itself.
How do you write pages AI answers can quote?
Write every section so it stays true, specific and attributed when lifted out alone. That’s the whole craft, and it’s what human skimmers want too.
A GEO pass on one page
Run it on your best page first.
-
Step 1 of 6
Match one heading to one question
Phrase each H2 the way people ask it, then answer in the next sentence. “How much does Semrush cost?” followed by the number beats a heading called Pricing.
-
Step 2 of 6
Make each section self-contained
Cut “as mentioned above” and “this tool”. Name the product, plan and number again, because the passage may travel without its neighbours.
-
Step 3 of 6
Swap adjectives for checkable facts
“Cheap” can’t be quoted. “$39 a month billed yearly, with 150 credits” can. Add the date you checked it.
-
Step 4 of 6
Name your sources in the sentence
“Google’s AI features guide says…” gives an answer engine attribution it can carry. A bare claim is easier to skip.
-
Step 5 of 6
Put comparisons in tables
Rows with matching columns are easy to retrieve and hard to misread.
-
Step 6 of 6
Keep facts fresh and consistent
Update prices when they change. Make sure your About page, product pages and outside profiles agree on the basics.
Looks like ordinary good writing? It is. GEO rewards the page that was already the clearest answer, and it punishes vagueness faster than Google ever did.
How do you measure GEO without guessing?
Start with the free reports from Google and Bing. Pay for a tracker only when you need to watch set prompts or report to a client.
Google Search Console has a Generative AI performance report, live worldwide since August 31, 2026. It shows AI Overviews and AI Mode impressions by page, country, device and date. It doesn’t show which queries triggered them.
Bing Webmaster Tools goes further. Its AI Performance report counts how often Copilot and Bing’s AI summaries cite each page. It also lists the grounding queries the AI used to find you, and a June 2026 update added Citation Share.
For ChatGPT, open your analytics. ChatGPT tags the links it sends with utm_source=chatgpt.com, so that traffic is easy to spot.
Paid trackers do what those reports can’t. They ask AI tools your chosen questions on a schedule, record who got named and show which rivals got cited instead, a job our competitor analysis tools roundup covers for classic search.
Heads-up: the Semrush, Surfer and RankPrompt links below are affiliate links. If you buy through one, we may earn a commission at no extra cost to you.
| Tool | AI tracking you get | Rough cost per check |
|---|---|---|
| Semrush Starter | 50 prompts daily, on top of the SEO plan | About 4 cents ($59.05/mo extra) |
| Surfer tracking plan | 50 prompts refreshed daily, mention gaps | About 5.5 cents ($82/mo, yearly) |
| RankPrompt Starter | 150 credits, each runs one prompt on 3 engines | About 26 cents ($39/mo, yearly) |
Those numbers aren’t like for like. One RankPrompt credit asks ChatGPT, Claude and Google AI Mode at once, so it’s closer to 9 cents per engine. We checked every price on the vendors’ own sites on 21 October 2026.
- Pick Semrush Starter if you already pay for its SEO plan, because $59.05 more a month adds AI tracking with the same SEO limits. Our Semrush pricing breakdown shows that premium across every plan.
- Pick Surfer if your team already writes in its editor, because tracking and optimizing then live in one place. Our Surfer SEO review covers the editor side.
- Pick RankPrompt for a cheap first read, because its free plan runs one full report across three engines with no card. Our RankPrompt review walks through what that report shows.
Whatever you choose, read trends, not single runs. The same prompt can get a different answer an hour later. A weekly batch of real buyer questions beats daily screenshots of one.
Frequently asked questions
Is GEO the same as AEO or LLMO?
Mostly, yes. AEO (answer engine optimization) and LLMO (large language model optimization) chase the same goal: being the source an AI answer uses. GEO is the term from the 2023 research paper, AEO is popular with tool vendors, and the tactics overlap almost completely.
How long does generative engine optimization take to show results?
Expect weeks, not days. OpenAI and Perplexity say robots.txt changes take about 24 hours to apply, but edited content only counts once the page is recrawled and answers are regenerated. Re-check your tracked prompts two to four weeks after a real update.
Can I pay to get my brand into ChatGPT answers?
Not into the answer itself, because organic citations can’t be bought. ChatGPT does show ads, but OpenAI labels them as sponsored and keeps them separate from the response. An ad budget buys a spot below the answer, not a citation inside it.
Does schema markup help with GEO?
It helps machines read facts like prices, authors and FAQs, but it isn’t a shortcut. Google says AI Overviews need no special schema, and your markup must match the visible text. Treat structured data as hygiene, not a citation trick.
What can I do when an AI answer gets facts about my product wrong?
Find the source it cited, because the error usually comes from a real page. Fix your own page first with the correct fact stated plainly. Then ask third-party sites to update stale details, and flag the answer with the tool’s feedback button.
Does publishing more AI-written content help GEO?
No. Volume without new facts gives an answer engine nothing worth quoting, and Google’s spam policies treat mass-produced pages as scaled content abuse. One page with original numbers or a real test earns more citations than ten rewrites of what already ranks.
How many prompts should a small site track?
Start with 25 to 50 questions a real buyer would ask, checked weekly. Include your brand name, your main product categories and two or three competitor comparisons. More prompts mostly add noise until someone on your team acts on the results.
The bottom line
Generative engine optimization isn’t a new discipline so much as a stricter grader. AI answers pick passages, not pages, and they favour the ones that state a checkable fact cleanly.
Rank on page one but rarely get cited? Run the six-step pass on your best page this week. Already own the top spot? Keep it fresh.
Either way, open the free AI reports in Search Console and Bing before you buy a tracker.
And start with the bot table. Thirty seconds with your robots.txt rules out the one GEO problem no amount of writing can fix.
