SEO gets your pages found and clicked in search results. GEO, generative engine optimization, gets your business named, quoted or linked inside AI answers. They share most of the work: readable pages, clear answers and facts you can prove. GEO adds crawler checks for each assistant, writing that is easy to quote, and separate measurement.
- GEO and SEO share most of the work. GEO adds crawler checks for each assistant, writing that is easy to quote, and measuring mentions apart from clicks.
- Assistants cite a small part of what they read: in the AIGROW Source Map, 19% of the sites read for an answer were cited in it.
- In the research that named GEO, adding citations, quotations and statistics raised visibility the most; adding keywords did not.
- Whether GEO pays is simple arithmetic: the monthly cost divided by what one enquiry is worth gives the enquiries you need.
GEO vs SEO: the short answer
SEO (search engine optimization) is the work of getting your pages found and clicked in search results. GEO (generative engine optimization) is the work of getting your business named, quoted or linked inside the answers AI assistants write. For a small business most of the work is shared: pages that crawlers can read, clear answers to buyer questions and facts you can prove.
GEO adds three things: checking that each assistant's crawler can reach you, writing that is easy to quote, and measuring mentions separately from clicks.
| Question | SEO | GEO |
|---|---|---|
| Where you appear | A ranked list of links | Inside a written answer, as a name, a quote or a source link |
| What you compete with | The other results on the page | Every page the assistant read, about 14 sites per web answer in the AIGROW Source Map |
| What decides it | The search engine's ranking systems | A search, then the model's choice of what to quote; with web search off, what the model already holds |
| Main levers | Indexing, relevance, links, page experience | The same, plus crawler access for each assistant, quotable facts and mentions on the sites assistants read |
| How you measure | Positions, clicks, Search Console | Share of answers that name you, citations, referrals from assistants |
| One check tells you about | One search engine | One assistant, one setting, one day |
AEO (answer engine optimization) is a third name for the GEO goal. Google's own guide says that, from Google Search's point of view, optimizing for generative AI search is still SEO. ChatGPT, Claude and Perplexity run their own crawlers and write their own answers, and that is where the extra GEO checks come from.
How AI assistants choose which sources to cite
With web search on, an assistant works like a quick researcher. It runs one or more searches, reads a set of pages, writes an answer and links some of them. Google calls the reading step grounding and the extra searches query fan-out: for "how to fix a lawn that's full of weeds", it may also search "best herbicides for lawns" and "how to prevent weeds in lawn".
With web search off, the assistant answers from what its model learned in training and links nothing.
The AIGROW AI Source Map recorded which sites ChatGPT and Claude read for 3,260 web answers in October 2026:
| What we counted | Result |
|---|---|
| Distinct sites read | 23,170, about 14 per answer |
| Sites read per answer by ChatGPT | 15.9 on average, over 2,580 answers |
| Sites read per answer by Claude | 6.6 on average, over 680 answers |
| Site reads that were also cited in the answer | 4,171 of 22,433, or 19% |
| reddit.com | Read in 71% of answers, cited in 2.8% |
Two lessons follow. About four in five site reads ended without a citation: the assistant read the page and left it out of the links in its answer. And assistants read different places: reddit.com was read in 89% of ChatGPT's web answers and in none of Claude's.
With web search off, the spread is wider. In the AIGROW Index, any two assistants shared 21% of the businesses they named. With web search on, 43% of the names were ones the model also gave without it. A plan built on one assistant's answers covers only that assistant.
What the GEO research found, and its limits
The term GEO comes from a research paper first posted in November 2023 and published at the KDD 2024 conference. Its authors rewrote source pages in nine ways and measured how much of each generated answer drew on them, across a benchmark of questions from 25 subject areas. In the paper's own words and figures:
| Change to the page | Result in the paper's test |
|---|---|
| Adding citations, quotations or statistics from credible sources | The three best methods: 30% to 40% more visibility on the main measure |
| Simpler or more fluent wording | 15% to 30% more visibility |
| A more persuasive, authoritative tone | No significant improvement |
| Adding more keywords from the question | Did not perform well |
Two more findings matter to a small business. Pages ranked lower in classic search gained the most: citing sources raised visibility by 115.1% for the page ranked fifth, while the same method cut the top-ranked page's visibility by 30.3% on average. And on Perplexity, a live answer engine, the methods raised visibility by up to 37%.
The limits are just as clear. The engines tested date from 2023 and 2024, the effect varied by subject, and the measure was a share of the answer text rather than enquiries or sales. The paper shows that presentation can change what an engine quotes.
It cannot tell you what one edit will do for your business this month.
How to write a page or blog post that earns AI citations
The research and Google's guidance point the same way: specific, sourced, first-hand content. In eight steps:
- Pick one question a buyer asks before paying, in their words. Your enquiry emails and calls are the best source.
- Answer it in the first two or three sentences, with the specifics: who, where, how much and how long.
- Back each claim with something checkable: a number with its date, a link to the official source, a quote from a named organisation, or your own records.
- Add what only you know: your process, how you set prices, jobs you have done, photos of your own work.
- Give each follow-up question its own heading, and answer it in the first sentence under that heading.
- Keep one strong page for each buyer decision. Google says pages made for every variation of a search, mainly to manipulate results, breach its scaled content abuse policy.
- Put a visible date on facts that change, such as prices and hours, and check them every quarter.
- Link the page from your service pages, and link out to the sources you quote.
Here is the difference on a hypothetical accountancy firm's service page:
| Version | Opening text |
|---|---|
| Before | We provide professional, tailored tax services for small businesses. |
| After | We prepare annual tax returns for sole traders and small companies in Leeds. The fee is fixed before we start and listed on our prices page. Once we have your records, a return takes about ten working days. |
The second version answers who, where, how much and how long, so an assistant has specific facts to quote when someone asks for an accountant for a sole trader in Leeds. Google's guide adds that you do not need llms.txt, special markup or content cut into small chunks. For crawler access and Google's requirements, see the AI search optimization guide.
What to do first: SEO, GEO or both
Start from the problem you can see:
| Your situation | Do this first |
|---|---|
| Key pages are missing from Google's index | Fix indexing and canonical URLs; the rest depends on it |
| You rank on Google, but assistants never name you | Check crawler access, then rewrite your main service page with the steps above |
| An assistant names you with wrong facts | Follow its source links and correct the fact where it is published |
| An assistant cites your article but names a competitor | Compare what the question asks for with what your page states |
| Assistants name you, but enquiries do not arrive | Test your contact or booking path on a phone |
| You have two hours a month | Improve one page a month and check one listing |
Split the effort between your own site and other sites. Your site carries real weight: in BrightLocal's September 2026 study of local searches, business websites made up 93% of the unique domains used as sources and 42% of all citations. More than half of the citations still went to other sites, and in the Source Map reddit.com alone was read in 71% of web answers.
Earn mentions on other sites the ordinary way: accurate listings, real reviews, trade bodies and local press. Google's guide warns that chasing inauthentic mentions is less helpful than it seems, as its systems are built to block spam. When you have real news, a press release puts a dated record of it on news sites; see how to distribute a press release.
Is GEO worth it? Work out the return
No published GEO return figure applies to your business, but the break-even point is one line of arithmetic:
enquiries needed per month = monthly cost ÷ (value of a new customer × margin × share of enquiries that buy)
A worked example with invented numbers for a small accountancy firm:
| Input | Example |
|---|---|
| Monthly cost of the work | $400, about six hours at $65, or an outside quote |
| Fees from a new client in the first year | $1,500 |
| Margin on those fees | 50% |
| Enquiries that become clients | 25% |
| Value of one enquiry | $1,500 × 50% × 25% = $187.50 |
| Enquiries needed to break even | $400 ÷ $187.50 = 2.1, so three a month |
Then count those enquiries. Add "an AI assistant such as ChatGPT, Claude or Gemini" to the "How did you hear about us?" question on your form and on calls.
Watch referrals in your analytics (OpenAI adds utm_source=chatgpt.com to links from ChatGPT), the Generative AI performance report in Search Console, and searches for your business name, since many people who see you in an answer look you up afterwards.
If after three months you cannot see enough enquiries to cover the cost, spend less or change the work. The guide on how to measure AI search visibility sets out the formulas for tracking mentions month by month.
Want to know where your business stands? The free checker shows whether ChatGPT, Claude, Gemini and Grok name it.
What paid GEO help should deliver
Pay for work you can inspect:
- a crawler and indexing check, with the exact changes made
- rewritten pages you approved, with a source for every claim
- corrected listings, with before and after
- a baseline of saved, dated answers you can run again
Before you agree on a scope, ask which questions will be tested, in which assistants, with which settings and how often. Turn down any offer that guarantees a ChatGPT position, promises to put your business into a model's training data or reports results it will not show you. Keep your accounts, pages and data in your own name.
The comparison of AI SEO tools covers tracking software, with prices checked in October 2026. Agencies that deliver this work to clients can use AIGROW's client seats and white-label services.
A 30-day plan to start GEO and test what gets cited
| Week | Work | What you keep |
|---|---|---|
| 1 | Write down ten buyer questions; ask each in two assistants, twice, and save the answers with the date | A baseline you can repeat |
| 2 | Check crawler access and indexing; compare your name, address and hours across your listings | A list of fixes, made |
| 3 | Rewrite the page behind your most common question, using the eight steps above | One page that answers plainly, with sources |
| 4 | Ask the same questions the same way; count the answers that name or cite you | A second reading to compare with the first |
For week 2, the AI crawler checker and the NAP consistency checker cover the mechanics. Read a change between week 1 and week 4 as a lead to follow up: models and sources also change on their own, so one month proves little. The free AI recommendations course has a checklist for keeping it going.
Where AIGROW fits
To get a first reading without doing the week 1 work by hand, the free AI scan derives five questions from your website, puts them to the models behind five assistants without web search, and counts the answers that name you, with a score out of 100. The AI Recommendation Rate page explains how that count is defined.
For a fuller baseline, the AI visibility audit covers 20 to 40 of your own buyer questions across four assistants. You get each answer dated and re-runnable, the domains the assistants cited and a list of fixes in order of impact. Neither the scan nor the audit can promise a mention.
Sources
Read on 20 October 2026:
- GEO: Generative Engine Optimization, KDD 2024 paper
- Google Search Central: optimizing for generative AI features on Google Search
- Google Search Central: AI features and your website
- BrightLocal: local AI visibility study, September 2026
AIGROW data: AI Source Map statistics, October 2026, AI Visibility Report, October 2026 edition and the Index method.
Check your own business in AI answers
Free GEO audit: see if ChatGPT, Claude, Gemini and Grok name your business, who gets named instead, and what your listings show.
Frequently Asked Questions
