Open Questions

By the end of October 2026, GEO and AEO supported software companies, agencies, and a growing body of research, as the earlier chapters of this history describe. Several basic questions were still open. This chapter takes five of them in turn: how visibility in AI answers is measured, what the practice should be called, how much of it depends on decisions made by a few platforms, who has the right to the content that AI systems use, and where optimization ends and manipulation begins. Each section starts from events that happened and reports what the sources say.

September 2025 to January 2026: Measurement

The tools described in The Agency Gold Rush rest on the assumption that visibility in an AI answer can be counted. Two events in this period tested that assumption from different directions. In September 2025, Google stopped supporting a search parameter, &num=100, that rank-tracking software used to retrieve a hundred results in one request. Semrush wrote on September 15 that its top 10 and top 20 data stayed fully intact,² and Ahrefs wrote on September 18 that the parameter no longer worked. Ahrefs restored top-100 tracking for enterprise customers on October 8 and for all keywords in Rank Tracker on November 14.⁹ A platform's decision about a technical detail had changed what the tools could report.

On January 28, 2026, Rand Fishkin published a SparkToro study in which 600 volunteers ran 12 prompts through ChatGPT, Claude, and Google's AI Overviews and AI Mode, for a total of 2,961 responses. The chance that ChatGPT or Google's AI would give the same list of brands in any two responses was below one in 100, and the chance of the same order was about one in 1,000. Fishkin wrote that a brand's percentage of appearances across many runs of the same prompt was probably a reasonable measure of its prominence, and that rankings were close to random.⁶

SparkToro Publishes a Study of AI Recommendation Consistency

Measurement of clicks, covered in Google Answers Back, and studies of which sources AI systems cite, covered in The Citation Economy, used other methods and measured other things than the two studies above.

2023 to 2026: What to Call the Practice

The vocabulary changed with the audience. The Origins of GEO shows that the exact phrase generative engine optimization appeared in one arXiv paper in 2023, none from January 2024 through June 2025, and 18 in each of two quarters of 2026, while only three papers spell out AEO in full, all from 2026.¹⁰ Exalt Growth's expert guide credits Jason Barnard of Kalicube with coining AEO years before GEO became common.¹¹ Wikipedia lists large language model optimization, artificial intelligence optimization, and AI SEO as other terms for the same practice.¹²

Companies chose different words in their own announcements. Microsoft Advertising's January 6, 2026, guide was titled as a guide to GEO.¹³ HubSpot described its agreement to acquire XFunnel in October 2025 in terms of AEO,¹⁴ and Omnicom Media Group's October 2025 report used GEO.¹⁵ Adobe's April 2026 release listed SEO, GEO, and agentic search optimization as separate solutions.¹⁶ In May 2026, Google's guide argued that optimizing for generative AI search is still SEO.¹⁷ The fuller account of the newest of these terms is in Agentic Search Optimization.

July 2025 to March 2026: Dependence on a Few Platforms

Several decisions by platforms between mid-2025 and spring 2026 changed what other businesses could do. Cloudflare began blocking AI crawlers by default on July 1, 2025, which changed which sites those crawlers could read.¹⁸ Perplexity stopped accepting new advertisers in October 2025, according to Adweek.⁴ In March 2026, OpenAI confirmed it was ending Instant Checkout in favor of retailer apps after Etsy, Walmart, and Shopify had lined up to sell inside ChatGPT, and Etsy told Modern Retail that the feature had not produced a large volume of sales.^19,20 Google's September 2025 parameter change, described above, altered what rank trackers could see.⁹

Each decision was announced by the platform and affected companies that had built products or budgets around the previous behavior. The reversals were not all in one direction. Google announced ads in AI Mode in May 2025 and new AI Mode ad formats in May 2026,^21,22 and OpenAI's ChatGPT Ads reached 89 countries by October 21, 2026, as described in Advertising Inside the Answer.²³

December 2023 to September 2026: Licensing and Copyright

The question of who may use published content to train and ground AI systems has two tracks: agreements and lawsuits. The Citation Economy covers the licensing agreements, including the Associated Press and News Corp agreements with OpenAI,^24,25 and The ChatGPT Shock covers The New York Times suit of December 27, 2023.¹ Google Answers Back covers the suit that a publisher filed against Google in September 2025.²⁶ TechCrunch reported in November 2024 that News Corp's Dow Jones and the New York Post had sued Perplexity.²⁷

Two later events show how far these disputes had come without a final answer on AI answers themselves. In September 2025, Anthropic agreed to a $1.5 billion settlement of a class action by book authors. The Authors Guild reported that about 500,000 titles qualified and that rightsholders could expect at least $3,000 per title before costs and fees. Judge William Alsup certified the class only for the alleged piracy of books and not for AI training, and the Authors Guild noted that his fair use ruling on training applies only to the three plaintiffs.³ On September 4, 2026, the parties in the New York Times case against OpenAI and Microsoft filed cross-motions for summary judgment on fair use before Judge Sidney Stein, and a ruling was still pending, according to AI Lawsuit Tracker.⁸ Cloudflare's pay per crawl beta of July 2025 offered publishers a third approach, which is a price per crawl.²⁸

April 2024 to February 2026: Optimization and Manipulation

The 2024 papers described in The Origins of GEO showed that text could be written to push language models toward a chosen product.^29,30,31 The commercial practice began to resemble those experiments in 2025 and 2026. On December 4, 2025, Ahrefs published a study of "best X" lists in ChatGPT's sources, and on December 12, Fishkin wrote that it showed how to effectively spam your way into AI mentions.^32,33

On February 10, 2026, Microsoft's security team reported that, over 60 days, it had found more than 50 distinct prompts from 31 companies in more than a dozen industries that tried to influence an AI assistant's memory. The prompts used "Summarize with AI" buttons and crafted links to instruct an assistant to remember a company as a trusted source or to recommend it first, and Microsoft called the technique AI Recommendation Poisoning.⁷

Microsoft Reports AI Recommendation Poisoning

Accuracy is a separate issue from manipulation. On October 22, 2025, the BBC and the European Broadcasting Union published a study in which journalists from 22 public service media organizations in 18 countries evaluated more than 3,000 responses from ChatGPT, Copilot, Gemini, and Perplexity. They found that 45 percent of answers had at least one significant issue, 31 percent had serious sourcing problems, and 20 percent had major accuracy issues, with Gemini worst at 76 percent.⁵

Google's guide treats optimizing for AI features as a form of SEO,¹⁷ while Microsoft's report labels the memory prompts as manipulation.⁷

Key Organizations