Insights · Article · November 16, 2026

GTM Advice I'd Push Back On: Inbound and Foundations.

Eight pieces of go-to-market advice founders keep hearing about AI — where each one breaks down, and what to do instead. Part one: inbound and the foundations underneath it.

Larry Chao ·

Two workers in hard hats and safety vests inspecting the hull of a large ship in a dry dock at dawn.

Most of the GTM advice founders hear about AI isn't wrong. It's applied too early, or on top of a foundation that can't carry it. This post covers the inbound side and the foundations underneath it; a second post covers outbound and what happens after the signature.

I've spent the past month rebuilding my go-to-market workshop for founders, reading what practitioners are publishing and testing it against what I see in the startups I work with. The same eight pieces of advice kept coming up. For each one, here is the common version, where I'd push back, and what I'd do instead.

1. "Buy the AI stack and you're AI-native."

AI scales whatever you already have. If your ideal customer profile is broad, your buyer is undocumented, and your value proposition fits everyone, an AI SDR, a forecasting tool, and a coaching tool will send that confusion to more people, faster.

The pattern shows up in the numbers. MIT's 2025 study of enterprise generative AI pilots found that about 95% delivered no measurable profit-and-loss impact. The methodology has drawn criticism, so I treat it as a warning rather than a precise rate. But it matches what I see: adoption is widespread, and very few teams can point to revenue it moved.

Kevin Patrick of Astris Partners described the failure well:

"Day 1 is when you see the automation. Day 30 is when you find out what you automated."

On Day 1, research runs on its own and follow-ups go out on time. By Day 30, sales finds the wrong accounts, signals that didn't mean anything, and meetings with no real intent behind them. The automation worked. The decisions underneath it didn't.

What to do instead: before buying another tool, write down four things. Which accounts are worth pursuing. Which signals make them worth pursuing now. The qualification bar a lead must clear before it reaches a person. And the context a handoff must carry so the next person can act. Then add AI to amplify what already works.

2. "Target a big market."

A big total addressable market is good for a pitch deck and bad for a first go-to-market. It describes who could buy, not who will buy this quarter or what problem they are trying to solve.

Maja Voje separates the ideal customer profile (ICP), the company you eventually want to serve, from the early customer profile (ECP), the customer who feels the pain now and will change a process for you. Early on, the ECP is the one that matters.

The difference is easy to hear. "Mid-market finance leaders" is a segment. "Finance teams that added new entities and struggle with month-end reporting" is a buyer someone can find, with a problem you can name in the first sentence.

What to do instead: keep the firmographics, and add the problem and the moment. Pick one use case, win it with a small group of similar customers, and only then widen.

3. "SEO is dead." Or the opposite: "Ranking first is all that matters."

Both are wrong in the same way: they treat search as one channel with one rule.

SEO is still the foundation. AI assistants still have to crawl the web, find your pages, and decide what they say. If your site blocks crawlers or hides its content behind scripts, you lose in traditional search and in AI answers at the same time.

But ranking is no longer the whole game. When a buyer asks an assistant, it often runs several searches behind the scenes and writes one answer from what it finds. Being named in that answer can matter even when nobody clicks. In one Ahrefs study, only 28% of mentions of Ahrefs in AI answers linked to its site. A separate Ahrefs study of 75,000 brands found a strong correlation (about 0.66) between how often a brand is mentioned across the web and how visible it is in AI answers. Correlation is not causation, but the direction is clear.

What to do instead: keep the SEO basics working, then add a layer on top. Be mentioned consistently across the topic, not just ranked for one keyword. Track mentions and links as separate numbers.

4. "Check your AI ranking."

There isn't one. Ask an assistant the same question five times and you can get five different answers. In one run your brand is named and your site is cited. In the next you are absent. In a third you are named but a reviewer's page is cited. In a fourth a competitor appears instead of you.

Will Scott of Search Influence put the question to a webinar audience this month: which of those runs is your ranking? None of them. AI visibility is a distribution, and one screenshot is one sample from it. Many founders find out about their AI visibility because an executive or investor sends a screenshot. That screenshot proves very little in either direction.

The assistants also differ from each other. Semrush's 2026 AI Visibility Index analyzed 126 million US prompts across four AI platforms and found only 36 brands stayed in the top 100 on all four, every month of the study. Doing well in one assistant tells you little about another.

What to do instead: pick 10 to 25 questions your buyers actually ask. Run each one several times in each assistant they use. Count mentions and citations as separate numbers, report each assistant separately, and compare yourself with your top three competitors. Start by hand or with free tools; a tracker for $30 to $100 a month is enough once you are publishing regularly. Treat the numbers like a windsock, not a GPS: watch the direction.

5. "Publish more, or find the trick that gets you cited."

There is a lot of advice about volume (thousands of AI-written pages) and tricks (special files, pages written only for AI agents). Neither addresses what the assistants are trying to do, which is to find an answer they can stand behind.

Will Scott's framework is useful here. A page has three jobs. It has to be retrievable: can the system reach it and tell what it is? It has to be extractable: can the system lift a useful answer from it? And it has to be corroborated: can the system verify the answer, and does the rest of the web say the same thing?

Two simple tests come out of that. First, the two-sentence test:

"If I copied only these two sentences, would the subject, claim, context, and evidence still be clear?"

Second, write one sentence that states who offers what to whom, such as "[Company] offers [product] to [buyer]," and use it everywhere: your homepage, directory listings, partner pages, press releases. When other sites describe you differently, AI systems have to guess which version is right. Inconsistent descriptions are an old SEO problem that matters more now.

What to do instead: put the direct answer in the first sentence of each section. Replace adjectives with facts a reader can check. Keep evidence next to the claim it supports. Then find where your facts should be confirmed elsewhere, such as directories, associations, and partner pages, and fix the places that conflict.

6. "Your company page is your LinkedIn presence."

Most startups put their LinkedIn effort into the company page. The data points the other way.

Meltwater analyzed 9.5 million AI citations across six assistants, using B2B prompts. 75% of the LinkedIn citations came from individual members' posts, and 25% from company pages. CEOs and founders were the most-cited roles, at 8.2% and 7.5% of cited content. Freshness mattered: 48% of the cited posts were under three months old, and only 12% were more than a year old. Ranked lists, comparisons, and decision frameworks were cited most.

A caveat: the study was published on LinkedIn's own marketing blog, so read it with that in mind. But it matches what a separate Meltwater report found: LinkedIn has become one of the most-cited sources in AI answers, especially in Perplexity.

There is a second benefit. People who comment on or like a founder's post are a signal of their own. Your audience can become a short list of people who already know your point of view.

What to do instead: have the founder, and the people who know the product best, post regularly under their own names. Write the comparison or checklist your buyer would ask an assistant for. Keep it specific: names, numbers, trade-offs.

7. "The website is a brochure."

Many startup websites are written to impress: a bold claim, a few logos, and a button to book a demo. Buyers, and the assistants that now answer for them, are looking for something else. They want enough to decide whether to take the next step.

A website that helps a buyer decide answers five questions. Who is it for? What work does it do? What changes for the customer? What proof supports the promise? What does deployment require? Then it makes the next step clear: a real workflow and its limits, how access and security work, and pricing or at least how buying works.

It also helps to think about each person who can say no. The person paying asks whether it is worth the money. The person using it asks whether it makes the job easier. The approver asks whether it is safe and compliant. A partner asks whether it fits what they already sell. Each of them may now ask an AI assistant first. A page for each question gives the assistant something to cite.

What to do instead: write one clear page for each buyer question, starting with the payer and the approver. Publish an FAQ with specifics: who founded the company, where it is, what the product does and doesn't do, and pricing. A thin public footprint leaves room for AI to fill the gaps with guesses.

8. "Inbound leads will wait."

Inbound is often treated as the easy part: the buyer came to you, so the hard work is done. The data says otherwise. In a Clay test of 6,346 demo request forms, two out of three never got a reply. The buyer raised a hand and nobody answered.

Speed is where AI earns its keep on inbound. Adam Schoenfeld's review of AI SDRs found the cold outbound versions mostly disappointed, while the inbound ones worked, because the job is bounded: qualify someone who already asked, using approved facts and a calendar, then hand off to a person. He cites ElevenLabs qualifying inbound leads in a median of four minutes, compared with two days by hand.

For a founder, the first version doesn't need a product. You are the best rep today. Write down how you qualify a lead, what disqualifies one, and what the first reply should say. That document becomes the agent's instructions later.

What to do instead: answer every inbound request within minutes, not days. Route each form fill to an owner with an alert and enough context to act. Measure the time from form to first reply, and fix that before buying more traffic.

Wrong too early, not wrong forever

None of this advice is wrong forever. Volume content, AI tools, and big markets all have their place. They are usually wrong too early, before the foundation can carry them.

You don't need to fix all eight at once. Find where your buyers get stuck and start there. For most early teams, that is one of three places: a value proposition that fits everyone, a website that doesn't answer the buyer's questions, or inbound leads that wait too long for a reply.

The next post covers the other half: outbound, and what happens after the signature.

Sources

By Larry Chao, Sovereign Pivot

← Back to Insights