Strategy
Five SEO myths, and the studies that end them
All five of these were sound advice when somebody gave it to you. That is what makes them awkward. Nobody sold your property a lie. They sold you the rules of a search engine that has since changed what it hands back, and the rules did not come with an expiry date on them.
So here they are, one at a time, each with the study that ends it. Where the evidence runs out we say so. The longer argument these are drawn from is in From SEO to GEO, which covers what changes for an independent property when an assistant writes the answer.
Myth one: rank first and you get found
Between 48% and 56% of the properties an AI assistant recommended did not appear anywhere in Google's organic results for the same question.
That comes from 695 unbranded hotel searches published by Tharro on 24 June 2026, run across the Algarve, Mallorca and Rhodes through ChatGPT and Google's AI Mode. The queries were ordinary ones: family hotels in Mallorca, where to stay in Rhodes. Roughly half the answers named properties that were not in the ranking at all.
Rank has not stopped mattering. The same study measured how often a property at each Google position got named.
| Position in Google's organic results | Named by an AI assistant |
|---|---|
| Top 3 | 66% |
| Top 10 | 61% |
| 11 to 20 | 45% |
| Below 20 | 27% |
Read that as a ladder rather than a gate. A property at position 15 still gets named nearly half the time. A property Google does not rank at all still turns up in answers, because the assistant built its list from sources that were never your website. Across those 695 searches a hotel's own domain was cited less than 10% of the time. The citations went to online travel agencies, review platforms and editorial guides, and a separate Cloudbeds study from June 2025 found OTAs alone accounting for 55.3% of all cited sources.
There is a second half to this myth, which is the assumption that some single list exists to climb. In the June searches ChatGPT named 3,151 distinct hotels and Google's AI Mode named 1,286. Only 691 properties appeared on both, an overlap of about 18%. Two assistants asked the same question about the same destinations returned substantially different worlds. There is no number one to aim at any more, which is a slower and less satisfying thing to be told than a rank.
Myth two: the traffic fell, so the SEO stopped working
Seer Interactive tracked 3,119 informational queries across 42 organisations between June 2024 and September 2025, covering 25.1 million organic impressions. Click-through on those queries fell from 1.76% to 0.61% where an AI summary appeared. The rankings behind them did not move.
Two other measurements point the same way. Ahrefs, working from December 2025 data, put the click-through drop for the top-ranking result at about 58% when a Google AI Overview appears. Pew Research Center ran browsing trackers on more than 900 US adults in March 2025 and logged 68,879 Google searches: people clicked a traditional result on 8% of visits where an AI summary appeared, against 15% of visits where none did.
None of that describes a site that broke. It describes a result page where the answer now sits above the links, and a reader who has what they came for before they reach them. A property can hold every position it ever had and watch the sessions fall, and this is the year a lot of owners will be shown that graph and told the SEO stopped working.
The same Seer analysis found the useful half. Brands cited inside an AI Overview earned 35% more organic clicks. The traffic is thinning and concentrating at the same time, on whoever gets named.
Myth three: work the keywords in
The Princeton GEO study presented at KDD 2024 tested what actually moves a page's visibility inside generative answers. Adding citations to sources raised it by up to 40%. Adding statistics raised it by 37%. Keyword stuffing lowered it by 10%.
Read the third number next to the first two. The tactic is not merely tired, it is subtracting. And the gains were largest for sites with the least existing authority, which is the one finding in this article that favours a small property over a group.
What the study rewards is close to what an editor would ask for. Say where a number came from. Answer the question you have raised. A page that quotes a source and gives a figure is a page an assistant can lift a sentence out of and attribute, and attribution is the whole transaction.
Myth four: schema markup is the secret
Google states that no special markup is required for its AI features. That is the company that owns the surface, saying so in its own documentation, and it settles the question of whether schema is the lever.
Other assistants do reward structured, extractable content, so schema is worth having. It is not the thing standing between a property and a recommendation. The work that looks like schema and is not is duller and matters more: getting the facts out of PDFs and booking widgets and onto pages. Rates, what is included, transfer times from the airstrip, the malaria position, whether the paths take a wheelchair, minimum stays, child policies. An assistant cannot quote a rate sheet it cannot open, and a great deal of what decides a booking is currently sitting inside a download.
Myth five: block the AI bots to protect your content
GPTBot, ChatGPT-User, PerplexityBot, ClaudeBot and Google-Extended are the crawlers behind the assistants doing the recommending. They are also the crawlers a robots.txt file blocks when somebody adds a protective line to it.
Plenty of properties have those agents disallowed without anyone having made the decision. It arrives in a template, or in a plugin's default, or with a well-meant note about protecting the photography. The effect is not protection. A blocked crawler cannot cite the property, so the property is not defended against the shortlist. It is missing from it.
Blocking is a legitimate choice for a business that has decided it does not want to be recommended by those platforms. It should be made on purpose, by a person, and written down.
What these studies do not cover
Every study quoted here looked at hotels. The Tharro searches ran in the Algarve, Mallorca and Rhodes. A four-night beach hotel booked six weeks out is not a tented camp booked eleven months out after four conversations, and we do not know whether these numbers hold at the top of the market.
Nobody has run this work on African safari lodges. Until somebody does, anyone telling you exactly how AI search behaves in this segment is guessing, ourselves included. What the five claims above have in common is not a prediction. Each of them is something somebody has now measured, and each measurement went the other way.
Common questions
Is SEO dead?
No. Ranking still correlates strongly with being recommended, and the technical work that helps a search engine also helps an assistant. What has changed is that ranking is no longer sufficient on its own.
Should we block AI crawlers to protect our content?
Only if you have decided you do not want to be recommended. Blocking GPTBot or PerplexityBot removes the property from those platforms' ability to cite it.
Do we need schema markup?
Google states that no special markup is required for its AI features. Other assistants do reward structured, extractable content. Schema is worth having and it is not the thing standing between you and a recommendation.
Which assistant should we optimise for?
None of them individually. ChatGPT and Google AI Mode named overlapping sets of hotels only about 18% of the time in the June 2026 study, so a property tuned to one is not thereby placed in the other. The conditions that make a property legible and credible are shared.
How would we know if this is working?
Ask the assistants directly, on a fixed set of queries, once a month, and write down what comes back. It costs nothing and it is more informative than any dashboard currently available.
The fuller argument, including where the assistants get their answers and why independent properties are hit hardest, is in From SEO to GEO: what AI search changes for an independent property.
Waeve is a performance marketing studio for luxury travel. We run the ads, build the sites and shape the sales process for a small number of properties each year.
Field notes go out as we write them. waevedigital.com
Sources
- Tharro, AI hotel visibility research, 24 June 2026. 695 unbranded hotel searches across the Algarve, Mallorca and Rhodes, run through ChatGPT and Google AI Mode. hotelnewsresource.com
- Cloudbeds, The Signals Behind Hotel AI Recommendations, 26 June 2025. Hundreds of automated queries across six destinations, 145 consistently top-ranked properties, via ChatGPT, Perplexity and Gemini. cloudbeds.com
- Seer Interactive, AIO impact on Google CTR, Tracy McDonald, 4 November 2025. 3,119 informational queries across 42 organisations, 25.1 million organic impressions, June 2024 to September 2025. seerinteractive.com
- Ahrefs, AI Overview click-through analysis, December 2025 data. medianama.com
- Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results, 22 July 2025. Browsing-tracker panel of 900+ US adults, 68,879 Google searches, March 2025. pewresearch.org
- Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024 (Princeton).
- Google, AI features and your website. developers.google.com
