GEO, short for generative engine optimization, is the work of making a brand easy for AI answer engines to find, understand and cite. A GEO audit tests how that work is going. This checklist has 20 checks in five areas, followed by five tests for choosing who runs the audit.
Pew Research Center found that 49% of U.S. adults use AI chatbots and that 60% read AI summaries at the top of search results, so these answers already shape how buyers meet a brand.
Last updated: October 17, 2026.
What is GEO, or generative engine optimization?
GEO, or generative engine optimization, is the practice of shaping a brand's pages, facts and third-party coverage so that AI answer engines such as ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews retrieve the brand accurately and name it in their answers. The word generative refers to engines that write an answer instead of listing links. Some teams call the same work answer engine optimization (AEO) or AI search optimization.
The goal is concrete. When a buyer asks an AI engine for the best option in a category, GEO aims to get the brand named, described correctly and linked to its own pages. The measurement behind that goal is the GEO audit.
How does GEO work?
GEO works on the pages and sources an AI engine reads before it answers. Some answers come from the model's training alone, and others are built from live retrieval: the engine searches the web, pulls passages from the pages it finds, and writes an answer that may cite a few of them. A brand can influence the second kind of answer by changing what those pages say and where they appear.
5W works on four levers. The table lists each lever and what it covers.
| Lever | What it covers | Why it matters |
|---|---|---|
| Answer-first pages | Question headers, a direct first sentence, named entities, self-contained paragraphs | An engine can lift a clear passage without the page around it |
| Consistent facts | Prices, specifications, names and dates that match across the site, listings and press materials | Mismatches produce wrong or hedged answers |
| Structured data | Schema markup that states facts in a form software can read | Crawlers read the facts without parsing prose |
| Third-party sources | Earned media, expert placements and reference sites that mention the brand | Independent pages corroborate what the brand says about itself |
How is GEO different from SEO?
GEO differs from SEO in what it targets: SEO aims for rank in a list of links, and GEO aims for a place inside the answer itself. The two overlap on technical basics, and the full comparison is in GEO vs SEO.
| Question | SEO | GEO |
|---|---|---|
| What is the goal? | A high position in search results | Being named and cited in an AI answer |
| What is measured? | Rankings, traffic, clicks | Appearance rate, first mention, citations, sentiment |
| Where does the work happen? | Mostly on the brand's own site | On the site and on the third-party pages engines retrieve |
| How is it tested? | Rank trackers and analytics | Fixed prompts run across several engines |
What is a GEO audit?
A GEO audit is a repeatable test of what AI answer engines say about a brand, and of the pages and sources behind those answers. It is the starting measurement for GEO, in the same way a technical review is the starting point for SEO.
The audit has one rule: test with fixed prompts and record the raw answers. A single question typed once tells you little, because answers change between engines and between runs.
What are the 20 checks?
The 20 checks run in five areas: prompts and baseline, visibility metrics, source map, owned site, and facts and cadence. Each table lists the check, the test and the pass.
Area 1: Prompts and baseline
| Check | How to test it | What a pass looks like |
|---|---|---|
| 1. Prompt set | List 50 to 100 questions buyers ask, across informational, comparison, recommendation and purchase-intent types | Every prompt traces to a real buyer question |
| 2. Engine coverage | Run each prompt in ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews | Results are recorded per engine, not blended |
| 3. Repeat runs | Run each prompt at least three times, because answers vary between runs | Every result shows a range, not one answer |
| 4. Competitor set | Add three to six named competitors to the same prompts | Brand and competitors are scored on identical questions |
Area 2: Visibility metrics
| Check | How to test it | What a pass looks like |
|---|---|---|
| 5. Appearance rate | Count the share of answers that name the brand | A number for each engine and each prompt type |
| 6. First-mention rate | Count how often the brand is named first | First-mention share tracked next to appearance |
| 7. Owned-domain citations | Count how often an answer links the brand's own pages rather than third-party pages | Appearance and owned citations reported as separate numbers |
| 8. Sentiment and positioning | Copy the lines that describe price, quality and who the product suits | Quoted lines filed by engine and prompt |
Check 7 matters because the two numbers can disagree. In 5W's wine data, DRC leads wine brands in AI answers at 36.9% but has 0% owned-domain citations. In 5W's test of 297 payments answers, Stripe led appearance at 80.5% while Airwallex led owned-domain citations at 44.3%.
Area 3: Source map
| Check | How to test it | What a pass looks like |
|---|---|---|
| 9. Top cited domains | List every domain cited across all recorded answers and rank them by count | A ranked list of the sources behind the category |
| 10. Brand presence on those domains | Check whether the brand appears on each top domain | Present and absent marked for every domain |
| 11. Source type mix | Sort the domains into news, review sites, forums, video and owned pages | The mix shown as shares by type |
| 12. Competitor source gaps | Find domains that cite competitors but not the brand | A gap list ranked by how often each domain is cited |
This area is the basis of the Citation Source Audit, which maps the publications, forums and platforms that engines cite for a category.
Area 4: Owned site
| Check | How to test it | What a pass looks like |
|---|---|---|
| 13. Crawler access | Read the robots.txt file for rules that block AI crawlers such as GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot and ClaudeBot | No unintended blocks on pages the brand wants read |
| 14. Readable HTML | View each priority page's source and look for the key facts in the HTML itself | Prices, specifications and answers appear without scripts running |
| 15. Answer-first pages | Read the first paragraph of each priority page | The direct answer comes first, including who the product suits |
| 16. Matching structured data | Compare each page's schema markup to the visible text | Schema describes only what the page shows |
Check 15 can move results by itself. In 5W's test of lip treatments, one added sentence stating who the product suits raised recommendation rates from 0% to 96.7% across ChatGPT, Claude and Gemini.
Area 5: Facts, earned media and cadence
| Check | How to test it | What a pass looks like |
|---|---|---|
| 17. Fact consistency | Compare prices, specifications, names and dates on the site, retailer listings and press kit | Every fact matches everywhere, or the mismatches are listed |
| 18. Independent coverage | Check for coverage from the last 12 months on the domains found in check 9 | Recent coverage on at least some top domains |
| 19. Error log | Record each wrong claim in an answer and the page that repeats it | Every error has a source and an owner for the fix |
| 20. Cadence | Rerun the full prompt set each month and compare to the last run | A month-by-month table tied to what was published |
Which findings should be fixed first?
Fix wrong facts first, then source gaps, then page structure. A wrong claim in an answer is the most direct damage, and checks 17 and 19 find it fastest.
Source gaps come next because they decide which pages an engine can retrieve about the brand. Page structure comes last. 5W's 2,400 controlled tests found that tables and bullets had little effect on whether ChatGPT, Claude or Perplexity used the correct information from a source, so accurate facts matter more than layout.
What mistakes ruin a GEO audit?
Four mistakes ruin most GEO audits: testing once, measuring only appearance, ignoring third-party sources, and treating layout as the fix.
- Testing once gives one answer, and answers vary between runs.
- Measuring only appearance hides the gap between being named and being cited.
- Ignoring third-party sources leaves out pages engines retrieve beyond the brand's site.
- Treating layout as the fix spends effort on tables and bullets when the facts are wrong.
How often should a GEO audit be repeated?
A GEO audit should be repeated monthly with the same prompts, because the comparison to the last run is the useful part. Track the same metrics each time, such as the six AI visibility KPIs communications teams should measure.
Results take repeated runs to show, because answers vary and no agency controls when or whether a model cites a brand. Compare each month's run to the last, and tie every change to the work that was done.
How do you choose the best GEO agency?
Judge a GEO agency on five tests: measurement, research, source access, control of the site and content, and proof before commitment. The table lists the question to ask for each test and how 5W answers it. A brand can also run a small version itself, with 10 prompts and three runs each in one engine, which covers checks 1, 2, 5 and 6.
| Test | Question to ask | How 5W answers |
|---|---|---|
| 1. Measurement | Does the agency test fixed prompts across engines and show you the raw answers? | The AI Visibility Audit runs 50 or more category prompts across ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews, maps the cited sources, benchmarks competitors and ends with a 90-day roadmap. |
| 2. Research | Has it published data with a method you can read? | 5W has published more than 50 reports in its AI Visibility Index, plus the 2,400-test study above. |
| 3. Source access | Can it place coverage on the sites AI engines cite? | 5W is a PR agency with its own media team, and its GEO services include earned media in trade publications. |
| 4. Site and content control | Can it change your pages and write the content, or does that wait for another vendor? | 5W does the work in house: media, site building and content, all of it. |
| 5. Proof before commitment | Will it test your brand before you sign? | 5W runs a free 10-prompt sample audit with no contract or retainer commitment. |
Why does doing the work in house matter for GEO?
Doing GEO work in house matters because a single fix usually needs three kinds of change at once: the brand's own pages, coverage on other sites, and new content that fills gaps. When three vendors own those pieces, each fix waits for a handoff and a new brief.
At 5W, the team that finds a gap in the audit is the team that edits the page, writes the content and pitches the reporters. Illustrative scenario: an audit shows engines cite a competitor's review roundup that leaves the brand out. The media team pitches that publication, the content team writes a comparison page, and the site team adds matching structured data from one brief.
5W's GEO Optimization Services cover six services: the audit and citation baseline, semantic content optimization, structured data and schema, third-party source development, brand entity cleanup and ongoing monitoring. Case studies for AI companies include Sahara AI and Dragontail Systems.
CONCLUSION
GEO, or generative engine optimization, comes down to one habit: test fixed prompts, record the answers, and trace every gap to a page or a source that can be fixed, ideally with a team that makes those fixes in house. Brands that want the 20 checks run for them can start with 5W's free sample audit and work with 5W on the full AI Visibility Audit and the fixes that follow.
