GEO Predictions for 2027: 3 Experts Share Their Takes

We sat down with three search experts and asked them what GEO looks like from here through 2027. They asked to be identified by first name only.

• Mike has spent years on the technical side of search, including how pages get retrieved and then used inside generative engines. He now spends most of his time on GEO programs.
• Jason has spent years on how engines understand a company as a brand, and on measuring whether that understanding is correct. He now watches the same problem inside AI recommendations.
• Lily researches how Google and other engines actually behave, including AI Overviews and what happens to clicks. Brands and publishers come to her for a reading of what changed.

Mike

We asked Mike what 2027 feels like for marketers making budget decisions.

It still feels a lot like the SEO meeting, honestly. I have done technical search work for a long time, so I know how those conversations go. The statement of work used to say we would get them onto page one for a list of terms. Now the same person is asking whether ChatGPT will recommend them, and whether the page is even in a shape a model can retrieve and use. And they of course still want an ROI expectation they can take to the C-suite.

The deals often start before anyone has written a statement of work. A product manager notices a competitor in Gemini, forwards the screenshot to procurement, and then Mike's phone rings.

A lot of 2027 deals will start that way. I used to tell people GEO would just fold into SEO, and then I saw what we actually signed, and it was GEO as a standalone service. However, internally, the work still lives with the SEO team, who is getting told to figure it out with a little online research. I expect next year to continue looking like that.

Even when the work is billed on an SEO invoice, Mike still treats it as a problem of getting pages retrieved and used by the model. The pages have to be written so a model can pull a passage from them. And the off-site coverage has to be strong enough that the model treats the company as a source worth using.

Predictably, RFPs are starting to ask for the same GEO / AEO criteria.

Clients used to ask for page-one keywords: a list of search terms that we'd have to rank on classic Google. The KPI I am seeing now, and what I think 2027 RFPs will require, is AI rankings and AI-generated commercial recommendations. It's about how often you show up when ChatGPT or Claude or an Overview recommends a company. They want to see that KPI reported more than once a quarter, and in more than one engine.

The better RFPs, Mike said, also ask whether the pages are written so a model can actually pull a passage: dates, definitions, comparisons, short specific claims. If the useful sentence is buried in paragraph 14 of a 3,000-word page, you can still win a classic ranking and lose the recommendation. The rest of the document is timelines and who is on the account.

Mike will buy tracking software: it shows where a brand currently appears in ChatGPT, Gemini, and AI Overviews, which is a cheap way to see the current picture. After that, people still have to do the work: rewrite the pages so a model can retrieve them, and build off-site coverage so the model treats the company as a source worth using.

What you are paying for is your position in ChatGPT, Gemini, AI Overviews, and the other engines that recommend companies. I would not tell a client we have this fully figured out, and anyone who says they have is selling certainty that is not there yet.

GEO becoming its own line on every invoice is not something Mike is counting on. Google, he thinks, will still be the main way people arrive at a vendor's site, even as chatbots take more of the questions. Connecting an AI ranking to a lead is the argument he still expects to be having in December.

Jason

Jason focuses on whether AI platforms characterize companies correctly. Often, being correctly defined is just as important as being more authoritative than competitors.

Does it know who you are, what you sell, and can it recommend you without mixing you up with a competitor? I check that across more than one engine, on a fixed list of commercial questions, every week if I can. Then I spend the rest of the meeting explaining that ChatGPT, Gemini, and Google do not all have the same picture. You can be described correctly in one place and wrongly in another, which is inconvenient if your VP only checks Google. I say that in meetings, and people nod, and then they go back and check Google anyway. I understand it: that is the habit, and it is just not a complete picture of whether you are being recommended.

We asked Jason what a company should take from the 2026 click research. If the recommendation is already on the results page, a lot of people never click through. Pew Research found that when an AI Overview showed up, people clicked a search result on 8 percent of visits, versus 15 percent when there was no Overview. A Washington University study on arXiv that spring crawled tens of thousands of trending queries and found almost 30 percent of the sites cited in those Overviews were not even on Google's first page.

Companies keep hoping Google's first page and the sites cited in AI Overviews will turn out to be the same list if they wait. The papers have not shown that. I still think of this as ranking: the ranking I get hired for is whether the engines recommend you, and whether they have your facts right when they do. Ranking the page does not fix a wrong description, and the facts have to match on the site and off the site.

Off-site proof, Jason said, is how an engine decides you are the company you say you are: journalists, partner pages, industry lists. Listicles show up in the logs. He has watched programs stall when the page the model would actually use is thin, or stale, or contradicts those sources.

People skip to the mentions because they are easier to screenshot, but I still start with whether the underlying description is right. Mention-counting tools are going to get cheap. By 2027 a Marketing Manager will be able to buy one without an approval, which is how it should be, in my opinion. They need to see a comparison of how their company is viewed across many dimensions and across all the relevant AI platforms. The measurement I want is whether you were described as the right company, with the right product.

On the invoice, Jason cares that someone owns the prompt list, and that someone is responsible for the brand being described correctly. Buried in a forty-page SEO retainer, that work gets looked at the month before the QBR. He prefers a dedicated engagement when he can get it, because the reporting has one job. Mike is right that both get signed. The bundled ones, in Jason's experience, are harder to hold to a ranking and to a correct description.

A ranking in one engine, Jason said, rarely shows up the same way in the others. He is not counting on a clean recommendation-to-revenue number in 2027, and he is not counting on the engines agreeing about who you are within a year.

Lily

Lily spends her time in the Overview data, and in the click data under it. Brands and publishers come to her for a reading of what actually changed. She is tired of decks that treat GEO like a brand-new category that needs a framework.

If you show me a framework with four quadrants I will probably end the call, because I have sat through that and it does not help me explain anything to a VP. What I actually need, and what I tell people they need, is a specific list of commercial prompts, the ones buyers type, and then some evidence that those recommendation numbers moved. I do not need a story about how the category is evolving, because I already see that every week in the Overview data.

A lot of what is being sold as GEO, Lily said, is SEO that already worked, with a new name on the statement of work. Ranking still feeds a lot of what the models pull. She wants people watching whether they get recommended. She also wants them to keep doing the Google work. Most visits to company sites still come from Google, and she does not believe chatbots become the main way customers arrive in 2027.

Inside companies it is still Google plus a Slack screenshot from someone on the team who asked ChatGPT a question over the weekend. That screenshot has had more power than the last two quarterly reports I have seen. I do not think that goes away in 2027, and if anything, more people inside the company will be asking ChatGPT those questions and forwarding what they see.

The teams that do this well, in Lily's experience, have one person on staff who owns the prompt list and the ranking in the recommendations, and they hire a shop for content and PR because they cannot staff that at the volume it takes. Mike sees a lot of GEO riding on SEO invoices. She thinks that is an accurate picture of how companies buy. If she were the client she would want someone on the account who can name the prompts.

I have been in meetings where the capability existed on a slide and nobody on the account could name the prompts, and I would not do that again. I have also watched companies buy a dashboard, like the way it looked, and get no movement in the recommendations, which is a bad way to spend a quarter. After kickoff I care about whether the ranking moved.

If you rank in Google and you are invisible when a buyer asks who to use, Lily said, that is a problem worth spending money on. GEO, as she talks about it, is ranking in the engines that recommend companies. When she has checked, that list and the classic Google list have not been the same.

Informational queries are where the click is getting thin, because the Overview can finish the job on the page. Transactional queries still send people through. That is the split she keeps seeing.

I will not buy a ninety-day thought-leadership sprint with no prompt list, and I will not buy PR that cannot say which commercial questions it is for. Listicles can help, technical work on the pages can help, I have seen both work. If we can name the questions and we can see the ranking move, then we can talk about 2028. The models still lean on search, and the companies I expect to still be getting recommended are the ones that keep doing good SEO and also watch the recommendation.

The three conversations kept returning to the same two facts. Companies are paying for position in the engines that recommend them, and Google is still sending most of the visits. Tracking software came up as something you might buy to watch the work. They did not agree on how new the work is, or on whether companies will buy it as its own engagement or as part of SEO.

References

• Chapekis, Athena, Anna Lieb, Sono Shah, and Aaron Smith. 2026. "Investigating Click Behaviors On Google Search Result Pages That Produce an AI Overview." arXiv:2608.04831. https://doi.org/10.48550/arXiv.2608.04831.
• Xu, Haofei, Umar Iqbal, and Jacob M. Montgomery. 2026. "Measuring Google AI Overviews: Activation, Source Quality, Claim Fidelity, and Publisher Impact." arXiv:2605.14021. https://arxiv.org/abs/2605.14021.