WEBVTT 00:00:00.000 --> 00:00:02.800 This is the Search Institute Podcast. 00:00:02.800 --> 00:00:05.560 This episode uses a synthetic narrator, 00:00:05.560 --> 00:00:08.600 and its content has been editorially reviewed. 00:00:08.600 --> 00:00:12.800 Is generative engine optimization really different from SEO? 00:00:12.800 --> 00:00:14.920 There’s a useful answer, but first 00:00:14.920 --> 00:00:18.440 we need to separate two things that often get mixed together. 00:00:18.440 --> 00:00:21.040 One is how a page gets found. 00:00:21.040 --> 00:00:24.120 The other is what happens when a system writes an answer 00:00:24.120 --> 00:00:25.960 using what it found. 00:00:25.960 --> 00:00:29.560 Google’s current guidance is direct about the first part. 00:00:29.560 --> 00:00:32.120 For AI Overviews and AI Mode, 00:00:32.120 --> 00:00:35.920 the technical and quality foundations of Search still apply. 00:00:35.920 --> 00:00:39.280 A supporting page has to be crawlable, indexed, 00:00:39.280 --> 00:00:43.200 eligible for a snippet, useful, and understandable. 00:00:43.200 --> 00:00:46.000 Google says there is no extra technical requirement 00:00:46.000 --> 00:00:49.080 for appearing as a supporting link in those features. 00:00:49.080 --> 00:00:52.360 But a generated answer creates a different outcome. 00:00:52.360 --> 00:00:55.200 A ranked results page orders documents. 00:00:55.200 --> 00:00:57.720 A generative system retrieves information, 00:00:57.720 --> 00:01:01.440 synthesizes prose, and decides which sources to cite, 00:01:01.440 --> 00:01:03.880 where to cite them, and how much of the answer 00:01:03.880 --> 00:01:05.640 each source supports. 00:01:05.640 --> 00:01:10.040 That’s where GEO, short for generative engine optimization, 00:01:10.040 --> 00:01:11.800 becomes a useful term. 00:01:11.800 --> 00:01:14.600 It extends SEO into the answer layer. 00:01:14.600 --> 00:01:17.720 It does not replace the search foundation underneath. 00:01:17.720 --> 00:01:19.800 Let’s start with that foundation. 00:01:19.800 --> 00:01:22.560 Google defines SEO as helping search engines 00:01:22.560 --> 00:01:25.680 understand content and helping people find a site 00:01:25.680 --> 00:01:27.600 and decide whether to visit it. 00:01:27.600 --> 00:01:29.400 Its guidance for generative Search 00:01:29.400 --> 00:01:31.360 keeps those jobs intact. 00:01:31.360 --> 00:01:34.840 AI Overviews and AI Mode draw from the Search index 00:01:34.840 --> 00:01:37.360 and use core ranking and quality systems 00:01:37.360 --> 00:01:41.040 to retrieve pages before a model assembles a response. 00:01:41.040 --> 00:01:43.120 Google’s checklist is familiar: 00:01:43.120 --> 00:01:47.120 allow crawling, make pages discoverable through internal links, 00:01:47.120 --> 00:01:50.120 put important information in indexable text, 00:01:50.120 --> 00:01:52.120 provide a good page experience, 00:01:52.120 --> 00:01:54.120 use helpful images and video, 00:01:54.120 --> 00:01:57.800 and make structured data match what people can actually see. 00:01:57.800 --> 00:02:00.080 Keep business and product information accurate 00:02:00.080 --> 00:02:01.400 where it applies. 00:02:01.400 --> 00:02:04.080 A page has to be indexed and snippet eligible 00:02:04.080 --> 00:02:06.760 before it can become a supporting link. 00:02:06.760 --> 00:02:09.960 Google has also rejected several AI-only rituals 00:02:09.960 --> 00:02:11.800 for its own Search products. 00:02:11.800 --> 00:02:14.320 Its July 2026 guidance says 00:02:14.320 --> 00:02:19.320 Search does not use an L-L-M-S dot text file as a special signal. 00:02:19.320 --> 00:02:22.800 It does not require pages to be broken into tiny chunks, 00:02:22.800 --> 00:02:27.360 prose rewritten just for machines, or a special GEO schema. 00:02:27.360 --> 00:02:30.120 Structured data still has its ordinary jobs, 00:02:30.120 --> 00:02:32.560 including rich-result eligibility. 00:02:32.560 --> 00:02:36.160 It does not create a separate pass into generative Search. 00:02:36.160 --> 00:02:40.000 That doesn’t mean every old SEO tactic is equally useful. 00:02:40.000 --> 00:02:42.520 Google emphasizes unique information, 00:02:42.520 --> 00:02:44.120 firsthand experience, 00:02:44.120 --> 00:02:47.880 clear organization, and content that helps a person. 00:02:47.880 --> 00:02:50.440 In the foundational GEO experiment, 00:02:50.440 --> 00:02:53.160 keyword stuffing performed poorly too. 00:02:53.160 --> 00:02:55.840 The shared ground is discovery, relevance, 00:02:55.840 --> 00:02:57.840 usefulness, and evidence. 00:02:57.840 --> 00:03:00.520 It is not a fixed checklist of tricks. 00:03:00.520 --> 00:03:02.160 Now move up one layer, 00:03:02.160 --> 00:03:04.960 from finding a page to writing an answer. 00:03:04.960 --> 00:03:07.640 In classic search, we can measure position, 00:03:07.640 --> 00:03:11.480 impressions, clicks, and what happens on a landing page. 00:03:11.480 --> 00:03:13.080 In a generated answer, 00:03:13.080 --> 00:03:15.800 a source might be retrieved and never cited. 00:03:15.800 --> 00:03:17.400 It might be cited late, 00:03:17.400 --> 00:03:19.000 used for one sentence, 00:03:19.000 --> 00:03:21.280 paraphrased with little link prominence, 00:03:21.280 --> 00:03:24.040 or replaced by another source in the next run. 00:03:24.040 --> 00:03:27.400 The research paper that introduced GEO, 00:03:27.400 --> 00:03:32.400 published at K-D-D in 2024, tried to measure this. 00:03:32.400 --> 00:03:34.560 Its authors used measures 00:03:34.560 --> 00:03:37.520 including position-adjusted word count 00:03:37.520 --> 00:03:41.960 and a broader impression score for citations inside an answer. 00:03:41.960 --> 00:03:45.720 The idea was to measure how much answer space a source gets 00:03:45.720 --> 00:03:47.920 and how prominently it is used, 00:03:47.920 --> 00:03:51.120 rather than simply assigning it another rank. 00:03:51.120 --> 00:03:55.840 A 2026 critical survey split the process into more stages: 00:03:55.840 --> 00:03:58.080 whether search activates at all, 00:03:58.080 --> 00:04:02.120 crawling and indexing, retrieval, reranking, 00:04:02.120 --> 00:04:04.120 space in the model’s context, 00:04:04.120 --> 00:04:07.440 citation, prominence, factual use, 00:04:07.440 --> 00:04:09.040 fidelity to the source, 00:04:09.040 --> 00:04:12.240 user behavior, and economic outcomes. 00:04:12.240 --> 00:04:16.560 One score called AI rank can hide several different failures. 00:04:16.560 --> 00:04:19.200 A page that never enters the candidate set 00:04:19.200 --> 00:04:22.040 needs a different fix from a page that is retrieved 00:04:22.040 --> 00:04:24.440 but loses the citation decision. 00:04:24.440 --> 00:04:27.760 That distinction changes the questions worth asking. 00:04:27.760 --> 00:04:29.800 Did the product search the web? 00:04:29.800 --> 00:04:31.520 Did it retrieve the page? 00:04:31.520 --> 00:04:33.200 Was the brand mentioned? 00:04:33.200 --> 00:04:35.200 Which URL was cited? 00:04:35.200 --> 00:04:38.280 What claim was the citation supposed to support? 00:04:38.280 --> 00:04:40.760 Was the source represented accurately? 00:04:40.760 --> 00:04:43.120 Did any of this lead to a qualified visit 00:04:43.120 --> 00:04:45.240 or a useful business outcome? 00:04:45.240 --> 00:04:49.000 A classic rank tracker cannot answer all of those questions. 00:04:49.000 --> 00:04:53.560 So how closely do AI citations follow ordinary search results? 00:04:53.560 --> 00:04:57.560 If every generated answer just cited the top ten blue links, 00:04:57.560 --> 00:05:01.760 GEO could be ordinary ranking work plus a new report. 00:05:01.760 --> 00:05:03.800 The evidence shows a relationship, 00:05:03.800 --> 00:05:06.160 but not a fixed transfer rate. 00:05:06.160 --> 00:05:10.960 In a 2025 analysis, Semrush sampled five thousand keywords 00:05:10.960 --> 00:05:14.560 and collected more than one hundred fifty thousand citations. 00:05:14.560 --> 00:05:18.880 It reported that eighty-two percent of Perplexity citation URLs 00:05:18.880 --> 00:05:22.000 overlapped Google’s top ten for the same query. 00:05:22.000 --> 00:05:25.720 For Google AI Overviews the figure was sixty-seven percent. 00:05:25.720 --> 00:05:29.520 For Google AI Mode it was about thirty-five percent. 00:05:29.520 --> 00:05:32.880 Picture three bars descending from roughly four-fifths, 00:05:32.880 --> 00:05:35.360 to two-thirds, to one-third. 00:05:35.360 --> 00:05:38.520 They describe different surfaces in one vendor study, 00:05:38.520 --> 00:05:40.920 not universal probabilities. 00:05:40.920 --> 00:05:44.400 Ahrefs studied fifteen thousand long-tail prompts 00:05:44.400 --> 00:05:47.160 with the corresponding Google and Bing results, 00:05:47.160 --> 00:05:51.200 plus visible citations or references from several assistants. 00:05:51.200 --> 00:05:55.680 It found an average of eleven point nine percent of those assistant URLs 00:05:55.680 --> 00:05:58.640 in Google’s top ten for the original prompt. 00:05:58.640 --> 00:06:02.080 Perplexity reached twenty-eight point six percent; 00:06:02.080 --> 00:06:07.000 other measured modes ranged from six point one to eight point six percent. 00:06:07.000 --> 00:06:10.400 Ahrefs’ separate work on Google AI Overviews 00:06:10.400 --> 00:06:12.480 found much higher overlap. 00:06:12.480 --> 00:06:16.600 Those Ahrefs and Semrush figures should not be averaged. 00:06:16.600 --> 00:06:20.640 They used different samples, interfaces, citation units, 00:06:20.640 --> 00:06:22.840 and ways of calculating overlap. 00:06:22.840 --> 00:06:26.360 They also should not be treated as a simple contradiction. 00:06:26.360 --> 00:06:29.560 Together, they say search ranking and AI citation 00:06:29.560 --> 00:06:30.720 are related, 00:06:30.720 --> 00:06:34.000 while the relationship varies by system and method. 00:06:34.000 --> 00:06:36.920 Other vendor datasets point in the same direction. 00:06:36.920 --> 00:06:40.280 Profound tracked six hundred eighty million citations 00:06:40.280 --> 00:06:42.680 and found different source-concentration patterns 00:06:42.680 --> 00:06:47.400 across ChatGPT, Google AI Overviews, and Perplexity. 00:06:47.400 --> 00:06:50.760 Frase later synthesized cross-platform research 00:06:50.760 --> 00:06:53.040 and argued for monitoring AI search 00:06:53.040 --> 00:06:56.800 as several retrieval systems rather than one channel. 00:06:56.800 --> 00:06:59.040 These are commercial measurement products, 00:06:59.040 --> 00:07:02.040 and neither reveals a stable ranking formula. 00:07:02.040 --> 00:07:05.040 Their useful warning is that the systems do not all draw 00:07:05.040 --> 00:07:06.760 on sources in the same way. 00:07:06.760 --> 00:07:08.440 What about experiments? 00:07:08.440 --> 00:07:12.840 Here the evidence is stronger, but the boundary matters even more. 00:07:12.840 --> 00:07:15.440 In the original GEO paper, researchers 00:07:15.440 --> 00:07:19.000 supplied sources to a benchmark generative engine. 00:07:19.000 --> 00:07:21.880 They rewrote one source using nine methods, 00:07:21.880 --> 00:07:23.800 then measured its citation impression 00:07:23.800 --> 00:07:28.240 across five random seeds and a ten-thousand-query benchmark. 00:07:28.240 --> 00:07:31.680 The best tested method improved position-adjusted word count 00:07:31.680 --> 00:07:33.560 by forty-one percent. 00:07:33.560 --> 00:07:36.400 The best method on the normalized subjective-impression measure 00:07:36.400 --> 00:07:39.320 improved it by twenty-eight percent. 00:07:39.320 --> 00:07:43.560 Relevant citations, quotations, and statistics performed well. 00:07:43.560 --> 00:07:45.640 Keyword stuffing did not. 00:07:45.640 --> 00:07:47.600 That result is meaningful. 00:07:47.600 --> 00:07:52.840 It is also narrower than saying GEO boosts visibility by forty percent. 00:07:52.840 --> 00:07:55.480 The experiment changed a source that was already 00:07:55.480 --> 00:07:57.000 in the engine’s context. 00:07:57.000 --> 00:08:00.120 It did not prove that a crawler would discover the page, 00:08:00.120 --> 00:08:03.120 that a live assistant would retrieve it for more prompts, 00:08:03.120 --> 00:08:07.120 or that the rewrite would produce durable traffic or revenue. 00:08:07.120 --> 00:08:10.440 A later controlled study made this distinction clearer. 00:08:10.440 --> 00:08:14.200 Vishwakarma and colleagues put exactly two candidate documents 00:08:14.200 --> 00:08:18.920 into six language models and ran two hundred fifty-two thousand paired trials 00:08:18.920 --> 00:08:21.320 across eighteen content factors. 00:08:21.320 --> 00:08:24.720 Topical relevance and position within the supplied context 00:08:24.720 --> 00:08:26.920 were the strongest drivers of which document 00:08:26.920 --> 00:08:28.920 received the first citation. 00:08:28.920 --> 00:08:32.000 Recent timestamps and explicit price information 00:08:32.000 --> 00:08:33.600 helped consistently. 00:08:33.600 --> 00:08:36.400 Formatting-only changes had little effect. 00:08:36.400 --> 00:08:38.760 This is strong evidence about competition 00:08:38.760 --> 00:08:41.120 among already-retrieved documents. 00:08:41.120 --> 00:08:43.880 It is not a test of open-web discovery. 00:08:43.880 --> 00:08:47.920 The 2026 critical survey reviewed forty-five studies. 00:08:47.920 --> 00:08:51.360 Its cautious conclusion was that changes to already-retrieved 00:08:51.360 --> 00:08:54.400 content can causally affect citation or use 00:08:54.400 --> 00:08:56.200 within tested settings. 00:08:56.200 --> 00:08:59.760 No reviewed technique established a stable, long-term, 00:08:59.760 --> 00:09:03.400 cross-platform causal effect on organic discoverability 00:09:03.400 --> 00:09:05.240 or downstream behavior. 00:09:05.240 --> 00:09:07.640 It also noted a competitive problem: 00:09:07.640 --> 00:09:10.200 if many sources adopt the same tactic, 00:09:10.200 --> 00:09:13.600 one source’s gain can shrink or shift elsewhere. 00:09:13.600 --> 00:09:15.560 Here’s a useful evidence ladder. 00:09:15.560 --> 00:09:18.920 First ask whether a page can be crawled and indexed. 00:09:18.920 --> 00:09:20.800 Then whether it is retrieved. 00:09:20.800 --> 00:09:24.600 Then whether an already-retrieved passage is used or cited. 00:09:24.600 --> 00:09:28.600 Finally ask whether a person visits, trusts, or acts. 00:09:28.600 --> 00:09:32.440 The experiments we just described mainly test the third rung. 00:09:32.440 --> 00:09:35.400 They do not prove every rung above or below it. 00:09:35.400 --> 00:09:38.440 That is why GEO deserves a workflow of its own 00:09:38.440 --> 00:09:41.960 only when it does work SEO is not already doing. 00:09:41.960 --> 00:09:45.080 Start with observations tied to a specific surface. 00:09:45.080 --> 00:09:50.280 For every test, record the prompt, locale, date, model or product, 00:09:50.280 --> 00:09:54.880 answer, brand mention, cited URL, citation position, 00:09:54.880 --> 00:09:57.240 and the claim the citation supports. 00:09:57.240 --> 00:10:00.920 Repeat prompts across runs because generated answers vary. 00:10:00.920 --> 00:10:04.680 A single screenshot is an example, not a visibility rate. 00:10:04.680 --> 00:10:07.320 Next, test answer-level hypotheses 00:10:07.320 --> 00:10:10.240 without calling them universal ranking factors. 00:10:10.240 --> 00:10:13.880 Original evidence, clear definitions, current facts, 00:10:13.880 --> 00:10:17.760 attributable claims, and concise answers to real subquestions 00:10:17.760 --> 00:10:19.880 can strengthen a page for readers. 00:10:19.880 --> 00:10:22.520 They may also help retrieval or use. 00:10:22.520 --> 00:10:24.760 Measure the effect on the specific surface 00:10:24.760 --> 00:10:26.520 before making a larger claim. 00:10:26.520 --> 00:10:28.680 Keep the whole funnel visible. 00:10:28.680 --> 00:10:33.160 Search Console and similar search data can show discovery and visits. 00:10:33.160 --> 00:10:36.520 Answer monitoring can show mentions and citations. 00:10:36.520 --> 00:10:39.800 Site analytics can show behavior after a visit. 00:10:39.800 --> 00:10:42.360 Each measure answers a different question. 00:10:42.360 --> 00:10:45.160 Citation share without qualified outcomes 00:10:45.160 --> 00:10:48.040 can be as misleading as rank without clicks. 00:10:48.040 --> 00:10:50.520 And inspect representation quality. 00:10:50.520 --> 00:10:52.760 A citation does not automatically mean 00:10:52.760 --> 00:10:55.160 the linked page supports the answer. 00:10:55.160 --> 00:10:58.040 Check whether the claim is actually on the page, 00:10:58.040 --> 00:11:00.360 whether the answer describes it faithfully, 00:11:00.360 --> 00:11:04.120 and whether a brand or piece of evidence is portrayed correctly. 00:11:04.120 --> 00:11:07.400 The critical survey warns that a citation can coexist 00:11:07.400 --> 00:11:09.160 with poor fidelity. 00:11:09.160 --> 00:11:11.880 Product behavior can also change quickly. 00:11:11.880 --> 00:11:14.360 The division of labor is straightforward. 00:11:14.360 --> 00:11:18.840 SEO owns crawlability, indexability, site structure, 00:11:18.840 --> 00:11:22.680 internal links, technical quality, page experience, 00:11:22.680 --> 00:11:25.800 search demand, and classic search measurement. 00:11:25.800 --> 00:11:28.920 GEO adds prompt and surface selection, 00:11:28.920 --> 00:11:32.200 answer capture, mention and citation measurement, 00:11:32.200 --> 00:11:35.000 representation review, repeated testing, 00:11:35.000 --> 00:11:37.320 and comparison across surfaces. 00:11:37.320 --> 00:11:40.200 Editorial work supplies the shared asset: 00:11:40.200 --> 00:11:42.200 original evidence that is current, 00:11:42.200 --> 00:11:45.720 attributable, well organized, and worth selecting. 00:11:45.720 --> 00:11:49.000 Before anyone sells you a new GEO tactic, 00:11:49.000 --> 00:11:50.680 ask four questions. 00:11:50.680 --> 00:11:52.840 What stage is it supposed to change: 00:11:52.840 --> 00:11:56.120 crawl, index, retrieval, citation, 00:11:56.120 --> 00:11:59.080 representation, click, or conversion? 00:11:59.080 --> 00:12:01.640 What primary guidance or experiment supports 00:12:01.640 --> 00:12:03.960 that mechanism on the target product? 00:12:03.960 --> 00:12:06.360 What stable denominator will show the result 00:12:06.360 --> 00:12:07.960 across repeated runs? 00:12:07.960 --> 00:12:10.120 And could the change harm usefulness, 00:12:10.120 --> 00:12:13.480 relevance, trust, or classic search performance? 00:12:13.480 --> 00:12:15.960 If a tactic has no stage, evidence, 00:12:15.960 --> 00:12:18.360 denominator, or failure condition, 00:12:18.360 --> 00:12:21.640 it may be a story attached to a changing interface. 00:12:21.640 --> 00:12:24.280 Google’s rejection of special AI files, 00:12:24.280 --> 00:12:27.240 artificial chunking, machine-only rewrites, 00:12:27.240 --> 00:12:30.280 and manufactured mentions is a useful reminder. 00:12:30.280 --> 00:12:33.640 A new output does not justify a new superstition. 00:12:33.640 --> 00:12:37.320 It is worth being precise about how this episode was researched. 00:12:37.320 --> 00:12:39.880 The source article compared three layers: 00:12:39.880 --> 00:12:41.960 the shared discovery infrastructure, 00:12:41.960 --> 00:12:45.400 the outputs SEO and GEO try to influence, 00:12:45.400 --> 00:12:47.800 and the measurements used to evaluate them. 00:12:47.800 --> 00:12:52.120 Its evidence, available through August twenty-fifth, twenty twenty-six, 00:12:52.120 --> 00:12:55.640 included four full-text Google guidance documents, 00:12:55.640 --> 00:12:57.800 three full-text scholarly papers, 00:12:57.800 --> 00:13:01.240 and four full-text vendor analyses or syntheses. 00:13:01.240 --> 00:13:04.360 This was comparative analysis of published evidence, 00:13:04.360 --> 00:13:07.080 not a new experiment by Search Institute. 00:13:07.080 --> 00:13:09.240 The definitions are operational, 00:13:09.240 --> 00:13:11.320 not an industry-wide standard. 00:13:11.320 --> 00:13:15.000 The percentages from Semrush and Ahrefs remain separate 00:13:15.000 --> 00:13:18.360 because their queries, interfaces, citation units, 00:13:18.360 --> 00:13:20.200 and calculations differ. 00:13:20.200 --> 00:13:22.760 Vendor studies can reflect commercial incentives 00:13:22.760 --> 00:13:25.320 and product configurations that change. 00:13:25.320 --> 00:13:28.760 The original GEO paper and the later competitive study 00:13:28.760 --> 00:13:32.520 appeared in peer-reviewed A-C-M conference proceedings. 00:13:32.520 --> 00:13:34.760 The critical survey was still a preprint 00:13:34.760 --> 00:13:36.440 at the research cutoff. 00:13:36.440 --> 00:13:39.480 No reviewed source proves that one content tactic 00:13:39.480 --> 00:13:42.680 causes durable discovery, traffic, conversion, 00:13:42.680 --> 00:13:44.920 or revenue across platforms. 00:13:44.920 --> 00:13:47.480 Google’s guidance describes Google Search, 00:13:47.480 --> 00:13:49.720 not every standalone assistant. 00:13:49.720 --> 00:13:53.080 Platform behavior can change after the article’s cutoff. 00:13:53.080 --> 00:13:56.360 Let’s close with the questions people usually ask. 00:13:56.360 --> 00:13:59.560 Is GEO just another name for SEO? 00:13:59.560 --> 00:14:00.280 No. 00:14:00.280 --> 00:14:04.360 They overlap, but GEO measures answer-level outcomes 00:14:04.360 --> 00:14:06.520 that a ranking report misses. 00:14:06.520 --> 00:14:11.160 Does Google require an L-L-M-S dot text file for AI Overviews 00:14:11.160 --> 00:14:12.520 or AI Mode? 00:14:12.520 --> 00:14:16.280 No, according to its July 2026 guidance. 00:14:16.280 --> 00:14:18.520 Such a file may have other uses, 00:14:18.520 --> 00:14:21.320 but it is not a substitute for crawlability, 00:14:21.320 --> 00:14:23.640 indexing, or useful content. 00:14:23.640 --> 00:14:26.840 Do Google rankings predict AI citations? 00:14:26.840 --> 00:14:29.960 Sometimes, unevenly, depending on product, 00:14:29.960 --> 00:14:32.600 query set, date, and definition. 00:14:32.600 --> 00:14:36.440 Did the original GEO study test organic discovery? 00:14:36.440 --> 00:14:37.080 No. 00:14:37.080 --> 00:14:40.280 It rewrote a source already supplied to an engine 00:14:40.280 --> 00:14:42.760 and measured answer-level impression. 00:14:42.760 --> 00:14:45.160 Can statistics and quotations help? 00:14:45.160 --> 00:14:48.280 In that controlled setup, relevant ones did. 00:14:48.280 --> 00:14:52.040 Unsupported numbers and decorative quotes are not a strategy; 00:14:52.040 --> 00:14:54.040 the evidence must be genuine 00:14:54.040 --> 00:14:57.480 and the effect retested on the intended surface. 00:14:57.480 --> 00:15:00.200 How should GEO performance be measured? 00:15:00.200 --> 00:15:02.920 Use a fixed and disclosed prompt set. 00:15:02.920 --> 00:15:06.120 Record the product, locale, date, response, 00:15:06.120 --> 00:15:10.760 mention, cited URL, citation position, supported claim, 00:15:10.760 --> 00:15:12.600 and how often it repeats. 00:15:12.600 --> 00:15:15.080 Keep those measures separate from impressions, 00:15:15.080 --> 00:15:18.120 clicks, engagement, leads, and revenue. 00:15:18.120 --> 00:15:20.600 Does this need a separate GEO team? 00:15:20.600 --> 00:15:22.360 Usually not at the start. 00:15:22.360 --> 00:15:26.840 Add these responsibilities to the people already working on technical SEO, 00:15:26.840 --> 00:15:30.600 content evidence, analytics, and brand representation. 00:15:30.600 --> 00:15:34.920 A dedicated team makes sense only if monitoring and experimentation 00:15:34.920 --> 00:15:37.960 become enough recurring work to require one. 00:15:37.960 --> 00:15:41.480 So, is GEO really different from SEO? 00:15:41.480 --> 00:15:45.320 Yes, at the level of output, measurement, and testing. 00:15:45.320 --> 00:15:49.560 A generated answer can retrieve, omit, cite, paraphrase, 00:15:49.560 --> 00:15:53.720 or misrepresent a source in ways a position report cannot show. 00:15:53.720 --> 00:15:56.680 But it is not a separate technical foundation. 00:15:56.680 --> 00:16:01.560 Google’s AI features still rely on crawlable, indexed, useful content 00:16:01.560 --> 00:16:03.560 and core Search systems. 00:16:03.560 --> 00:16:08.040 Controlled studies show that already-retrieved content can affect citations. 00:16:08.040 --> 00:16:11.160 The broader evidence does not yet give us a durable formula 00:16:11.160 --> 00:16:14.040 for open-web discovery or business impact. 00:16:14.040 --> 00:16:17.080 The practical model is SEO underneath, 00:16:17.080 --> 00:16:20.600 GEO on top, and evidence at the center. 00:16:20.600 --> 00:16:22.680 Build material worth finding. 00:16:22.680 --> 00:16:24.840 Keep the search foundation sound. 00:16:24.840 --> 00:16:28.760 Then measure how each generative surface selects and represents it.