← Back to Blog

How to Get Your Brand Cited in AI Search: The TACT Framework

By Swati Paliwal, Co-Founder at ReSO ·

Your next buyer will probably decide whether to shortlist you before they ever land on your website. They will ask ChatGPT, Perplexity, or Google's AI for the best options, read the answer, and move on. If your brand is not named in that answer, you are not in the running.

This session broke down what actually changed in search, why keyword optimization no longer gets you cited, and the four-part framework Swati Paliwal uses to get brands surfaced inside AI answers. It also covered the working audit and content system her team runs for clients week over week. Here is the full playbook.



What actually changed in search?

Search used to return a list. Now it returns an answer.

That single shift explains the pattern Swati calls the crocodile mouth effect: impressions keep climbing while clicks fall away. The gap between the two lines widens like an open jaw. Platforms want people to stay and read the answer, not click out, so the AI resolves the question in place.

For the first time in a decade, Google faced real competition. In the first quarter of 2025 it lost search market share to ChatGPT and Perplexity, then responded with AI Overviews, which absorbed the old knowledge panel, and later AI Mode. The way brands earn visibility changed the moment both the platforms and buyer behavior changed at the same time.

The practical consequence is that the buyer now runs the entire journey, top to bottom, inside the LLM. Awareness, research, and the final shortlist all happen in the conversation. Swati cited research putting the zero-click rate around 68% of the time, meaning most AI answers never send the buyer to a website at all. If you are still optimizing only for the top of the funnel, you miss the buyer at the exact moment they decide.

Why keyword optimization stopped working

Search moved from a keyword ecosystem to a prompt ecosystem.

Type the same words into ChatGPT from two different accounts and you get two different answers. The engine reads intent and context, not just the string of words. Older search treated a dictated paragraph and a typed phrase the same way because it matched keywords. AI treats them differently because it is answering the meaning behind the words.

The mental model that follows: buyers are no longer searching, they are asking. They treat the LLM as an expert and expect a recommendation, the way you would ask a knowledgeable friend. A single prompt usually bundles several intent signals at once, so you cannot keyword-match your way into the answer. You have to match intent.

What are the two prompt categories buyers use?

The funnel is the seller's view of the buyer. Intent is the buyer's own view of their purchase. Swati's framework maps the funnel onto the intent behind each prompt, and sorts every prompt into two categories.

Informational prompts sit at the top of the funnel and break into three intents:

Intent What the buyer is asking
Strategy How do I do this? What is the best way to approach it?
Risk How do I avoid this? What are the common mistakes and the cost of getting it wrong?
Operational How do I rate this? What are the KPIs that define success?

Brand-inducing prompts sit at the bottom of the funnel and matter most, because this is where the buyer uses AI to validate a decision and forces it to name a brand. There are six intents here:

Intent What the buyer is asking
Discovery What are the best options?
Feature and capability Who actually does this specific thing?
Comparison to alternatives What are the alternatives to Company A?
Persona and use case What works best for someone like me?
Pricing and transactional What does it cost?
Geography Are you present and compliant in my market?

Brand-inducing discovery is where the sale is won or lost. If the buyer is about to sign with a competitor and asks the LLM for alternatives, that answer decides whether you even get a look. Swati shared a case where a single negative Reddit thread was the only source AI could find for competitor comparisons, so it fed every comparison answer. The fix was not clever, just absent until then: publish real content, case studies, LinkedIn and YouTube posts that give AI other sources to pull from.

The persona intent has a hidden trap. When someone asks "what works best for someone like me," the LLM already assumes the context from their history, so a startup founder and an enterprise CMO get completely different answers to the identical prompt. That is why serious prompt testing has to specify the persona and run through APIs, so the results are clean and not skewed by your own account history.

How do the AI platforms differ?

The top layer is the same everywhere: prompt plus user context, then retrieval, then live search, then the model synthesizes an answer. What differs is the underlying data.

ChatGPT, Perplexity, Gemini, Claude, and Bing Copilot each pull from a different dataset. That last one matters more than teams expect: many enterprises block ChatGPT on company machines and allow only Bing Copilot, so an enterprise ICP may be living in a platform you are not even tracking.

Google holds the freshest index on the planet because it crawls and ingests continuously. ChatGPT works from an existing dataset it updates on its own cadence. That difference drives strategy. A new company will find it easier to break into Gemini and Perplexity first, while ChatGPT takes longer to pick you up. An older company with a five-year-old blog is often better served by refreshing that existing content than by starting from scratch, since the domain history is already working in its favor.

The TACT framework

TACT is the four-part formula for getting cited. It comes down to one core idea: who you say you are has to match what the internet says about you. When those two line up, AI trusts you and surfaces you. When they diverge, you drop out of the answer.

Letter Pillar What it covers
T Technical optimization Machines can read you as well as humans can. Mobile readiness is essential; sites that are not mobile-optimized do not get cited. Do not block crawlers on interlinked content. Meta descriptions matter for Perplexity and Google, less so for ChatGPT.
A Answer-ready content Content mapped to intent, not to who you are. Structured, specific, easy to consume. Content that only describes your company gets cited far less often than content that answers a question.
C Citations What the internet says about you. Quality of the sites linking to you matters more than a domain authority score. Toxic backlinks with high spam scores actively hurt.
T Trust and proof of entity Alignment between your positioning and third-party signals. If reviews and backlinks describe you as the agency you used to be rather than the product company you are now, the mismatch keeps you out of answers.

A few pointed clarifications came out of the discussion:

Domain authority is overrated for AI. A brand-new site with one genuinely smart content piece can get cited, while a site with a respectable authority score and toxic backlinks struggles. The quality and relevance of who links to you outweighs the number.

Do not fake refresh dates. Adding fresh dates to page HTML worked until roughly a month ago, when Google's latest update explicitly warned against it. Programmatic pages published at scale that all carry the same date send a bad signal too. The safe move is to add a genuine line or two of new content, or interlink to another piece, rather than rig the timestamp or rewrite the whole page.

Which citation sources actually move the needle?

Reddit sits at the top of the citation list again and again. It is also the one channel you cannot automate.

Any account under 500 karma gets banned, and if an agency flags your brand with accounts that get banned, it can take three months to rebuild Reddit's trust even with authentic activity afterward. The team does this manually and slowly for clients: no clicking shared links to threads, because clicking tags the account and gets it banned. They share screenshots instead. If you want to build presence yourself, spend time posting about whatever you personally enjoy until you clear 500 karma, then start on anything professional.

Wikipedia is the other high-value source, but only if you work with an editor who has real experience in your industry. An editor outside your space, however credible they sound, has a much lower chance of getting the page to stick. That expertise requirement is the trust-and-proof pillar showing up again.

One more principle that AI rewards: original data beats repackaging. The old SEO move of combining the top three ranking articles into one comprehensive piece does not work anymore. AI is hungry for new, structured, unique information, so one well-researched primary-research article per quarter, interlinked across your other blogs, will start lifting those other pages into citations too. Publicly available product documentation helps for the same reason, because it answers specific questions the exact way an LLM likes to consume them.

How the audit and content system works

Swati's team packages all of this as an agentic employee rather than another dashboard to learn, because the CMOs and founders she spoke with did not want one more tool to log into. The team takes ownership of the work and reports results back.

Here is what a run looks like, and what any team can replicate manually:

  • Prompt selection. The client picks their prompts from a shared library, typically around 50, weighted toward brand-inducing intents (for example 35 brand-inducing and 15 informational), plus their own prompts. The agent scopes the universe from the client's site, digital assets, and competitor assets.
  • The citation math. To answer roughly 40 to 50 queries, AI pulls from 400 to 500 domains and analyzes more than 1,000 links, around 1,200 URLs on average. Those URLs are the map. Where a competitor's URL is the source and yours is not, that is a gap you can close.
  • Reporting. For each selected prompt you see whether you appear, your rank, and your visibility share across ChatGPT, Perplexity, and Google, plus where competitors appear and you do not. Benchmarking against roughly 300 companies that show up in AI answers isolates your true direct competitors from the enterprise names that are not really in your bracket.
  • Weekly cadence. Everything runs weekly. Running daily produced random variance, while weekly showed real movement. Each week the team ships content, landing-page fixes, and new backlinks together, so strategy is paired with execution.
  • The fix list. Landing pages get checked against 200-plus variables, from meta descriptions to structure.

Prioritization is lowest-hanging fruit first: brand-inducing prompts where competition is thin and the revenue impact is real, because early movement in pipeline is what makes the work stick. From there the team weighs where visibility already exists, whether the prompts you rank for actually match your ICP, and where competitors are pulling ahead.

Key Takeaways for GTM Teams

  • Optimize for the bottom of the funnel, not just the top. The buyer now completes the whole journey inside the LLM, and brand-inducing discovery prompts are where you get shortlisted or skipped.
  • Match intent, not keywords. A single prompt carries several intent signals. Map your content to what the buyer is trying to accomplish.
  • Run TACT as a system. Technical readability, answer-ready content, quality citations, and trust alignment work together. A gap in any one keeps you out of answers.
  • Audit which URLs AI actually cites. Pull the sources behind the prompts that matter to your ICP, find where a competitor is the source and you are absent, and create the content that closes that gap.
  • Reddit and Wikipedia reward patience and expertise, not automation. Build karma slowly, work with in-industry Wikipedia editors, and never let an agency automate your way to a ban.
  • Publish original data. Repackaging the top-ranking pages no longer works. One strong primary-research piece a quarter, interlinked across your blog, lifts everything around it.

Want to Attend the Next Session?

The GTM Clawllective is a community of go-to-market operators sharing how they build with AI in real time, with live marketing and sales sessions every week. Recaps like this one and the speakers' resources live in the Collective's Resource Hub. If you want to catch the next session live and ask your own questions, join the community.


About the Speaker

Swati Paliwal is the co-founder of ReSO, where she builds AI-powered generative engine optimization and GTM systems that get brands cited inside AI search. She previously worked at Disney+ and has spent the last two years developing this approach, starting with playbooks tested across clients in late 2024 and refining it into the framework and audit system she runs today.

Her work focuses on the shift from keyword search to AI-mediated discovery, and on the practical mechanics of earning citations across ChatGPT, Perplexity, Gemini, and Google AI.

Watch the session replay: How to Get Your Brand Cited in AI Search: The TACT Framework