How to Do GEO: Step-by-Step AI Visibility Guide (2026)
AI assistants are becoming the first stop in B2B buying. A procurement manager no longer types “best lead management tool with CRM integration” into a search box. They ask ChatGPT or Gemini and get one synthesized answer with a few named vendors. That is why the question marketing teams ask has changed. Teams now want to know how to do GEO and how to tell whether the work actually paid off. This guide goes beyond definitions. It lays out a closed-loop process that starts and ends with measurement.
Short answer: You do GEO (Generative Engine Optimization) in four moves. First, identify the questions your customers ask ChatGPT, Gemini, Perplexity and Google AI Overviews, and measure whether your brand appears in the answers. Next, give AI crawlers technical access, publish clearly defined, evidence-backed content and add structured data. Then earn a place on the third-party sites AI engines read. Finally, re-measure every week.
What is GEO, and why is “how to do GEO” a process rather than a task?
GEO stands for Generative Engine Optimization. In Turkish it is called Üretken Motor Optimizasyonu. It means optimizing your brand and content so that AI answer engines mention, recommend and cite you.
The goal is not a ranking position. You win in GEO when three things happen inside the AI answer:
- Your brand is mentioned.
- It appears near the top of any list.
- Your site is linked as a source.
Scope note: This article covers GEO as AI search optimization. It is not about geographic data (geo-targeting) or geometry, two meanings that often get mixed up in search results.
GEO is not a one-off project. AI answers shift over time, models get updated and competitors publish new content. The work is therefore a loop: measure → fix → re-measure.
What is the difference between GEO, SEO and AEO?
| SEO | AEO (Answer Engine Optimization) | GEO | |
|---|---|---|---|
| Goal | Ranking in organic results | Winning the featured answer (snippet, voice) | Being mentioned and cited in AI-generated answers |
| Success metric | Position, clicks, CTR | Featured snippet / PAA ownership | Mention rate, citation rate, share of voice |
| Signal source | Links, content, technical health | Concise Q&A formatting | Search index + third-party mentions + entity clarity + training data |
| Measurement difficulty | Low (Search Console) | Medium | High (answers vary on every run, no official console) |
Does GEO replace SEO? No. Most AI engines answer commercial questions with real-time web search. A page that is poorly indexed or slow to crawl will not be retrieved, and a page that is not retrieved cannot be cited. GEO is built on top of SEO.
Training data vs. real-time retrieval: A language model knows the web as it was up to its training cutoff date. When the engine needs fresh information, it runs a search, reads the top pages and writes an answer from them. This is called retrieval-augmented generation (RAG). You cannot quickly change what a model learned in training. You can, however, influence the retrieval layer within weeks: which pages get found, how quotable they are and which third-party sites describe you.
Step 1: How do you measure your current AI visibility before you start?
Without a baseline, you cannot answer “did GEO work?” Measure first, then change things.
Manual method:
- Ask each prompt in every engine at least 3 times.
- Use a private window or logged-out session to limit personalization.
- Log every result in a spreadsheet.
AI answers are non-deterministic. The same question can name different brands in different runs, so a single test is misleading.
Record 5 criteria for every answer:
- Mentioned? (yes/no)
- List position (1st, 2nd, 5th…)
- Cited? Is your site linked as a source?
- Tone: positive, neutral or negative
- Accuracy: are prices, features, location and founding facts correct?
Sample baseline template:
| Prompt | Engine | Mentioned | Position | Cited | Tone | Accurate |
|---|---|---|---|---|---|---|
| best lead management software for B2B | ChatGPT | Yes | 4 | No | Neutral | Price outdated |
| best lead management software for B2B | Perplexity | No | – | – | – | – |
| lead management software pricing | AI Overviews | Yes | 2 | Yes | Positive | Correct |
Then run the same prompts for 3-5 competitors. Your share of voice is your mentions divided by the total mentions of every tracked brand.
Step 2: Which questions does your audience ask AI? How do you build a prompt set?
Group prompts by customer journey stage. Here is an example for a B2B lead management software company:
Awareness
- What is lead management software?
- How do I choose a lead management tool?
- What is lead scoring and how does it work?
- How do I follow up with leads faster?
- What is WhatsApp marketing automation?
Consideration
- Best lead management software for B2B companies
- Can you recommend a lead tool with WhatsApp integration?
- Lead management software vs. CRM: which should I use?
- Best sales automation tools for small teams
- Which lead tool works with Turkish-language WhatsApp?
- Alternatives to spreadsheets for tracking leads
Decision
- Lead management software pricing 2026
- Affordable lead management tool for startups
- Reliable sales automation company in Istanbul
- Is [Brand] good for B2B teams?
- [Brand] reviews and complaints
- Which lead tool has a free trial?
Where to find prompts:
- Questions your sales team hears on calls
- Google “People Also Ask” boxes
- Customer reviews
- Long, question-style queries in Search Console
Prompts are longer and more conversational than keywords. Cover phrasing variants as well. Turkish users, for example, write “en iyi”, “tavsiye” and “önerir misin” for essentially the same intent.
Step 3: Technical accessibility: can AI crawlers read your site?
Check robots.txt for these user agents:
| User agent | Owner | Purpose |
|---|---|---|
| GPTBot | OpenAI | Collects training data |
| OAI-SearchBot | OpenAI | Indexes pages for ChatGPT search results |
| ChatGPT-User | OpenAI | Fetches a page when a user's request requires it |
| PerplexityBot | Perplexity | Builds Perplexity's search index |
| ClaudeBot | Anthropic | Collects training data |
| Google-Extended | A control token for Gemini training and grounding; it does not affect Google Search or AI Overviews |
The key distinction is between training bots and search bots. You may block training bots for policy reasons. If you block OAI-SearchBot or PerplexityBot, however, you cannot appear as a source in those engines.
User-agent: OAI-SearchBot
Allow: /
User-agent: ChatGPT-User
Allow: /
User-agent: PerplexityBot
Allow: /
Silent blocking: Robots.txt is not the only gate. CDN and firewall bot-protection settings can return 403 errors or challenge pages to AI crawlers without anyone noticing, and some providers now block AI crawlers by default. Check your server logs for these user agents.
JavaScript rendering: Some AI crawlers do not execute JavaScript. Prices, product descriptions and FAQs should be present in the server-rendered HTML.
llms.txt: This is a proposed Markdown file at your domain root that gives language models a curated map of your most important pages. robots.txt tells crawlers what they may access. llms.txt suggests what is worth reading.
# Example Software
> Lead management and sales automation software for B2B teams.
## Product
- [Features](https://www.example.com/features): Lead capture, scoring, follow-up
- [Pricing](https://www.example.com/pricing): Plans and 2026 prices
## Resources
- [FAQ](https://www.example.com/faq)
Honest note: No major engine has officially confirmed using llms.txt as a ranking or citation signal. Treat it as a low-cost extra, not a core step.
SEO basics checklist:
- Pages are indexable (no stray noindex tags).
- Canonical tags are correct.
- The XML sitemap is up to date.
- Core Web Vitals are acceptable.
- Internal linking reaches key pages.
Step 4: How do you write content that AI engines cite?
Build semantic blocks. The first 1-2 sentences of every section should answer the heading's question directly, so the block can be quoted on its own.
Increase evidence density. Use statistics, cited sources and expert quotes. The GEO study by Princeton and collaborators (Aggarwal et al., 2023) found that adding citations, quotations and statistics can boost source visibility in generative engine responses by up to roughly 40%.
Use structured formats. Comparison tables, clear definitions, numbered steps and FAQ sections all make content easier for AI engines to extract.
Before/after example:
Before: “In today's fast-changing world, businesses need innovative solutions to manage their customers, and our platform offers many benefits.”
After: “Lead management software collects inbound leads from forms, ads and WhatsApp into one list, scores them by intent and triggers follow-up automatically. According to [source], responding within the first hour sharply increases qualification rates.”
The second version has a definition, a mechanism and a verifiable claim. Every claim in it can be quoted.
Add dates and freshness. Include prices, years and a “last updated” line. Outdated information comes back in AI answers as wrong information.
Write decision-stage pages. Pricing, comparison and “who is this for” pages answer the prompts closest to revenue. To connect that visibility to pipeline, see how marketing teams turn campaign demand into qualified pipeline.
Step 5: How do you clarify your brand entity and structured data?
AI engines think in entities. Your brand name, what you do, your category, your location and your founders should be described identically across your site, LinkedIn, directories and press. Open your About page with a one-sentence definition: “X is an Istanbul-based SaaS company that builds lead management software for B2B teams.”
Use these schema types where they fit:
- Organization
- Product/Service
- FAQPage
- LocalBusiness
- Article
Use sameAs to link your LinkedIn page, Wikidata entry (if eligible), Google Business Profile and social accounts.
{
"@context": "https://schema.org",
"@graph": [
{
"@type": "Organization",
"name": "Example Software Inc.",
"url": "https://www.example.com",
"description": "Istanbul-based SaaS company building lead management software for B2B teams.",
"sameAs": [
"https://www.linkedin.com/company/example",
"https://www.wikidata.org/wiki/Q000000"
]
},
{
"@type": "FAQPage",
"mainEntity": [{
"@type": "Question",
"name": "Who is Example Software for?",
"acceptedAnswer": {
"@type": "Answer",
"text": "B2B sales and marketing teams of 5-200 people."
}
}]
}
]
}
Entity SEO, semantic SEO and structured data matter because they reduce ambiguity. A model that is unsure what your brand is will skip it rather than risk an error.
Step 6: Off-site sources: which sites does AI read before it answers?
Open the source links in the AI answers you logged in Step 1 and count which domains repeat. Those are the authorities the engines trust in your category.
In Turkey, recurring source types include:
- Turkish “en iyi X” (best X) listicles and industry roundups
- News sites
- Şikayetvar (complaints platform)
- Ekşi Sözlük (community forum)
- Google Maps reviews
- Sector directories
- Reddit and niche forums
Ways to get included:
- Digital PR
- Pitching list authors
- Systematically collecting reviews
- Supplying data to comparison articles
A common reason AI gives wrong information about a brand is an outdated third-party page, such as an old price on a directory or a discontinued feature in a review. Find these pages and request corrections.
How do ChatGPT, Gemini, Perplexity and Google AI Overviews choose sources?
| Engine | Search infrastructure | Citation style | Watch for |
|---|---|---|---|
| ChatGPT | Its own web search feature | Inline links and a sources panel when search is used | OAI-SearchBot access; third-party mentions |
| Gemini | Grounding with Google Search | Links shown in some answers | Google indexing; Google-Extended setting |
| Perplexity | Its own index and Sonar models | Numbered citations in nearly every answer | PerplexityBot access; fresh, quotable pages |
| Google AI Overviews | Google's search index | Link cards beside the summary | Classic SEO; featured snippet and PAA presence |
Appearing in one engine but not another is normal, because each engine uses a different index and different retrieval logic. Measure each engine separately. For AI Overviews, classic ranking signals matter most. Pages that already win featured snippets or PAA boxes are frequently cited.
Step 7: How do you track results? KPIs and a 90-day GEO plan
KPI definitions:
- Mention rate: Answers mentioning you ÷ total answers. Example: 20 prompts × 4 engines × 3 runs = 240 answers. You are mentioned in 60, so your mention rate is 25%.
- Average list position: The mean rank when you are mentioned.
- Citation rate: Answers linking your domain ÷ total answers.
- Share of voice: Your mentions ÷ mentions of all tracked brands. Example: 60 ÷ 300 = 20%.
- Tone distribution: The percentage of positive, neutral and negative mentions.
- Accuracy: The share of mentions with correct facts.
Re-measure weekly. Answers fluctuate, so a trend across several weeks means more than any single snapshot.
| Period | Focus | Output |
|---|---|---|
| Days 0-30 | Baseline measurement, robots/CDN/rendering fixes, prompt set | Baseline report, technical fix list |
| Days 31-60 | Semantic blocks, evidence, decision pages, schema, entity consistency | 5-10 updated or new pages |
| Days 61-90 | Off-site sources, corrections, reviews, PR; re-measurement | KPI comparison, new task list |
Analytics: Create a channel group for AI referrers such as chatgpt.com, perplexity.ai and gemini.google.com. AI Overviews clicks arrive as Google organic traffic and cannot be separated this way. Many AI mentions produce no click at all, so visibility measurement must sit alongside traffic data. Tie both to pipeline in your analytics view.
How do local businesses and SMBs do GEO?
Local buyers ask “city + category” prompts, such as “best accounting firm in Ankara” or “reliable dental clinic in Kadıköy.” For these prompts, the Google Business Profile, review volume and quality, and local directories carry heavy weight. Gemini and AI Overviews in particular draw on Google's local data.
Minimum 5-step GEO for a tight budget:
- Complete and verify your Google Business Profile.
- Ask every satisfied customer for a review and reply to all reviews.
- Write one page answering your top 10 customer questions, with prices and dates.
- Check that robots.txt and your CDN do not block AI crawlers.
- Run 5 city + category prompts in 3 engines every month and log the results.
Should you do GEO yourself, use a tool or hire an agency?
Manual measurement adds up fast. 20 prompts × 5 engines × 3 runs equals 300 checks every week. At about a minute per check, that is roughly five hours of logging before any analysis.
| Option | Cost | Time | Best for |
|---|---|---|---|
| Manual | Free | High (hours per week) | Very small prompt sets |
| Tool | Low monthly fee | Low | Teams with in-house content capacity |
| Agency / managed service | Higher | Minimal | Teams without GEO resources |
What to look for in a tool:
- Which engines it measures
- The 5 criteria per prompt
- Competitor comparison
- Source-site analysis
- Turkish support
- Automatic weekly runs
Jumpix AI Visibility is one example. It covers:
- Engines: ChatGPT, Gemini, Perplexity, Google AI Overviews and optionally Claude.
- Scoring: Mention, position, citation, tone and accuracy for every prompt.
- Competitors: Share of voice against up to 5 competitors.
- Next steps: A task list and content briefs generated from the gaps it finds.
- Price: USD 9.99 per month for one brand.
- Free options: An instant check with no signup (industry + city, 5 questions, 3 engines) and a card-free trial.
- Language: A Turkish-language dashboard.
Teams without in-house capacity can use the managed GEO service at jumpix.app/ai-visibility/geo-hizmeti.
Conclusion
GEO is a measurable discipline, not a checklist you complete once:
- Measure your baseline across engines.
- Build prompts from real buyer questions.
- Fix technical access, quotable content, entity clarity and off-site sources.
- Re-measure weekly.
Brands that run this loop consistently are the ones AI engines learn to recommend.
Frequently Asked Questions
What is GEO work and how long does it take to show results?
GEO work is the process of making AI engines mention and cite your brand. Technical fixes and new content can affect real-time retrieval within a few weeks. Off-site authority and entity signals typically take 2-3 months. Training-data effects take longer still, because they depend on model updates.
What does GEO do, and what does it bring to my business?
GEO puts your brand inside the answers buyers read when they ask AI for recommendations, often before they ever visit a website. Its benefits include earlier brand presence in the buying journey, better qualified inbound demand and correction of inaccurate information AI may spread about you.
Does GEO replace SEO? Should I do both?
No, GEO does not replace SEO. Most AI engines rely on real-time web search, so pages that are not well indexed rarely get cited. Run both: SEO secures retrieval, and GEO makes your content and brand easier to quote and recommend.
How can I check whether my brand appears in ChatGPT or Gemini answers?
Ask your key prompts in each engine at least three times in a private window. Log mention, position, citation, tone and accuracy for every answer. For weekly tracking across many prompts and competitors, an automated AI visibility tool saves hours.
Is uploading an llms.txt file enough for GEO?
No. No major engine has officially confirmed that it uses llms.txt as a citation or ranking signal. It is a cheap extra, but crawler access, quotable content, structured data and third-party mentions matter far more.
Related reading
<!-- rank-agent-related -->- geo aracı <!-- rank-agent-link:0f0a3a00-77c7-414c-8bcc-8920fc5d37e2 -->
- geo danışmanlığı <!-- rank-agent-link:51fdc314-7483-4413-827d-5fb645743dac -->
- SEO vs GEO: Comparison Table, Measurement and 2026 Strategy <!-- rank-agent-link:b6690d4f-01d3-48d4-9742-2d7af5fd192d -->