Answer engine optimization, and designing a site for humans and agents at once: Linda Tong of Webflow

Answer engine optimization is the newest acronym in a field that generates them faster than anyone can settle on one. Linda Tong's contribution is a more useful framing than the acronym: your website now has two audiences with genuinely incompatible requirements, and nobody has decided how to serve both.

Tong is CEO of Webflow, the visual development platform, with around 300,000 customers. She joined as COO and president, running sales, marketing, education, support, people operations, finance and legal, before moving into the top job. Before Webflow she was CEO at AppDynamics, VP of Product and Innovation at the NFL, chief product officer at Tapjoy, and in product at Google.

Why AI is an accelerant rather than a threat here

Webflow's stated vision has always been to bring developer capabilities to everyone, which in its current form means building websites visually.

Tong's point is that this places the company in an unusually fortunate position. For many companies AI offers productivity improvements. For Webflow it is a direct accelerant to the vision itself, because there is no technology better suited to putting the power of code in more hands.

Paired with what she calls their core differentiator, being visual-first so people can tweak and interact with what they build, she describes it as a match made in heaven.

The consequence for planning is the interesting part. Webflow used to talk about unlocking capabilities incrementally, on a three, five or ten year horizon, because without AI it would genuinely have taken that long. Her assessment now is that the timeline has compressed sharply and they will reach the long-term vision far sooner.

What changed inside the company

Michael's own example sets it up: he built a functional application in 48 hours for about a hundred dollars, work that would previously have cost a fortune and taken over a year.

Tong describes gains on both sides. Individually, people are generating code, writing and reviewing test cases, fixing errors, and building context around a complex platform. Elsewhere in the business, content creation and the automation of functions that were genuinely not possible before.

Her specific example is Webflow running hundreds of experiments on its own website, automated through the optimization product that came with an acquisition the prior year.

But the framing she keeps returning to is not replacement. She describes AI as a partner that removes drudgery and unlocks work she would never have attempted.

Her illustration: designing multi-variate tests across thousands of elements on a website. Previously impossible to do manually. Now a system can suggest what to test, propose the variants, create them, run the tests and select winners.

Michael's observation is the right one. Even a simple two-variant test used to require multiple developers and significant time, and half the time you finished without knowing what to do with the result. That cost has largely disappeared, which makes it possible to take things that were static and make them dynamic.

The surprise

Asked what surprised her most, Tong names two things.

The pace of model quality improvement, measured in days or weeks between releases, which has fundamentally shifted what can be created. This against a backdrop where plenty of people were resistant, arguing the output was not good enough, and where the early wave of low-quality generated content made people feel under siege.

And then the ways people are actually using it, which she finds more surprising than the obvious ones of drafting documents and writing code.

Her own example is from the day of the recording, her three-year anniversary at Webflow. She fed her historical documents, email and calendar into a model and asked for a summary of her three years.

Her point is that this is work she would never have done, because it would have been prohibitively time-consuming. And she found the summary accurate and genuinely enjoyable to read.

What she took from it is worth noting for anyone who has never stopped to look back. She has never taken a break between jobs and had never truly reflected. Sitting with three years of accumulated work let her feel some pride in what had been accomplished, and was humbling, because none of it was individual.

Two audiences, one website

The core argument, and the reason this episode matters beyond Webflow.

Tong believes there needs to be either a standard or a new practice, because the two audiences want different things.

Websites made for humans are built for experience, beauty, layout, structure and design.

What an agent wants is to efficiently understand, index and source the right content, so it can surface it accurately in a chat result or use it to reason.

And her verdict on the current state: the way sites are designed today, an agent scraping one is highly inefficient. She thinks the industry needs a new standard for how an agent navigates a site, interacts with it, and ultimately takes action through it.

Her diagnosis of why nobody has solved it is honest. People are too busy trying to get one good website out to think about building two.

The compounding difficulty is that as traffic splits between the two sources, companies are trying to optimize a single site for both. And ranking in a language model is fundamentally different from search engine optimization, so blending the two has unpredictable effects on volume from either.

Her framing of what Webflow is studying: how might the web look when a site genuinely has two versions.

The acronym problem, and what to actually optimize for

Michael and Tong exchange the acronyms in circulation, AEO, GEO, AIO, with answer engine optimization the most prominent.

The example they both reach for is Stack Overflow, whose traffic collapsed as models absorbed the answers.

Tong's advice on how to think about it is refreshingly specific, because her answer is that the question is different for every company.

Take a startup whose website exists to drive signups into a product, or a straightforward commerce site where you want someone to sign up and buy. Getting the same traffic back is not necessary, because visitors arriving through a model may already be further down the funnel. What matters is surfacing what you sell effectively inside the assistant, while maintaining some organic search presence, and getting people to conversion faster. Less traffic, more conversion.

Which is a genuinely different objective from a site whose entire value is volume of eyeballs on user-generated content, where the traffic itself is the product.

Her implicit instruction: work out which of those you are before deciding whether falling traffic is a problem.

What comes after creation

The most forward-looking part of the conversation is Tong's account of what the current tooling wave has not solved.

She has built applications across every tool in the category, and names the thing nobody discusses. You have created fifty or sixty or seventy applications and spent a couple of hundred dollars. What are you doing with them?

The gap she identifies is the lifecycle. Continuously managing these things, iterating on them, adding value, sharing them. Because without ongoing purpose, curation and iteration, they become, in her phrase, internet garbage.

So her definition of the next phase is doing that at scale, and building the tools that let people take an idea they have turned into something and grow it into something meaningful.

Which produces a neat statement of ambition. Webflow has talked about owning the full web lifecycle. She now wants to own the full code lifecycle.

The 5 things I took away from this conversation

1. Decide which audience your site is actually for. Human visitors want experience and design. Agents want efficient indexing and clean structure. Those are different products, and most companies are quietly trying to serve both with one artifact and wondering why neither works well.

2. Falling traffic is not automatically a problem. Linda's framing is the most practical advice I have heard on this. If your site exists to drive conversion, arriving visitors may already be further along, so fewer of them converting better is a win. If your value is eyeballs on content, it is not. Work out which you are before panicking.

3. Multi-variate testing at scale is now actually available. The reason nobody ran hundreds of experiments was not that it was a bad idea. It was that a two-variant test needed developers, time and enough traffic to reach significance. That cost has largely gone, which means static things can become dynamic.

4. Ask for the retrospective you never make time for. Linda fed three years of documents, email and calendar into a model and asked for a summary of her tenure. That is not a productivity trick. It is a piece of work nobody does because the cost was prohibitive, and it apparently produced something worth reading.

5. Creation is solved, lifecycle is not. Everyone can now generate applications quickly and cheaply. Almost nobody has an answer for what happens to them next. That gap is where Linda thinks the next phase of tooling lives, and she is probably right.

FAQ

What is answer engine optimization? The practice of making content discoverable and citable by AI assistants and answer engines rather than only by traditional search. Tong notes it goes by several acronyms and that ranking inside a language model works differently enough from search engine optimization that the two can conflict.

How do you optimize a website for AI agents? Tong's view is that no adequate standard exists yet. Sites built for human experience are inefficient for an agent to parse, and she argues the industry needs a new practice for how agents navigate, interact with and act through a site, potentially meaning two versions of the same site.

Should companies worry about losing search traffic to AI? It depends what the traffic was for. Tong's distinction is between a site that exists to drive conversion, where visitors arriving through an assistant may be further down the funnel and fewer visits can still mean more signups, and a site whose value is volume of visits to content, where the loss is direct.

What does AI change about running experiments on a website? It removes the cost that limited them. Previously a simple two-variant test required developer time and produced results that were hard to act on. Tong describes systems that suggest what to test, generate variants, run the tests and select winners across thousands of elements.

What is the full code lifecycle? Tong's term for the problem after creation. Tools now make it trivial to build applications, and there is no equivalent answer for maintaining, iterating on, curating and growing them. She frames owning that lifecycle as Webflow's next ambition.

Also mentioned

• Webflow, its visual-first platform and its acquisition of an optimization product
• AppDynamics, the NFL, Tapjoy and Google, the earlier stops in Tong's career
• Stack Overflow, whose traffic collapse is the reference case for AI-driven search change
• The competing acronyms for optimizing to AI assistants, and why AEO leads
• Prezi and Hex, where Tong holds board seats

Listen to the full episode

Linda Tong on Between Two COO's

Between Two COO's is hosted by Michael Koenig. Subscribe on Apple Podcasts, Spotify, or wherever you listen.