SHARE
SHARE
In Part 1, we covered what changed in AI search and the foundational fixes every founder should make first. This part goes deeper: how to research what people are actually asking AI, how to optimize your content for citations, how to build authority beyond your own site, and how to measure whether any of it is working.
Once again, we’re drawing on Lily Grozeva‘s expertise to guide us through the tactics. If Part 1 was the why and the where to start, Part 2 is the how.
Prompt Research: The New Keyword Research
Keyword research told you what people searched. Prompt research tells you what they ask AI – and whether you show up in the answer.
Start with a manual audit before optimizing anything. Run a few queries across ChatGPT, Perplexity, and Google AI Mode – something like “What are the best [category] tools for [your ICP]?” See if your brand appears, how it’s described, and who it’s grouped with. This is your baseline.
Then build a targeted prompt list:
- 10 non-branded prompts – category-level questions your ICP asks
- 10 branded prompts – direct questions about your company, pricing, integrations, alternatives
The best source for these isn’t an AI-generated list. It’s your actual sales calls, support tickets, and Reddit threads where your ICP discusses the problem you solve. Real customer language gives you far better precision than anything AI will suggest.
One caveat: as of early 2026, there’s no data on actual prompt volume the way there is for Google keywords. Let your ICP research, core offering, and existing website content guide your selection.
"The ultimate goal is to maximize visibility across your targeted prompts."
Lily Grozeva
Founder Saltanat Labs
Content AI Loves and Cites
Once you know which prompts matter, make sure your content earns citations for them. Quality, facts, and credibility matter far more than volume. AI models extract specific facts, trade-offs, and consensus to build their answers, not just the most comprehensive page.
👉 Start with the pages AI is already citing. Open any of the AI visibility tools out there (ex. RankScale, Peec, Profound), run a report for your targeted prompts, and filter the results to your domain only. This shows you which of your pages AI models are already pulling from when generating answers.
These are your highest-leverage assets. The connection is already established, so optimizing them yields faster results than creating something from scratch. Aim to identify your top ~20 and start there.
Content formats AI regularly cites:
- “Best X for Y” comparison lists
- “Alternatives to [Competitor]” pages with genuine analysis
- Pricing pages with specific numbers
- Technical documentation that honestly covers trade-offs
- Case studies with measurable, specific results
When optimizing those pages, replace marketing language with facts. “Fast deployment” becomes “avg. deployment: 10 days.” For comparison pages, a consistent structure works best:
Criteria → Definition → Measurement → Where you win/lose.
Honesty about limitations actually increases credibility with AI.
A few more things that matter:
- Prioritize bottom-of-funnel pages (comparisons, reviews, pricing) over generic blog posts
- Date-stamp your content and refresh key pages every 90 days
- Add original visuals with descriptive captions and alt text
And one thing to resist: rewriting your entire website from scratch. Start with the pages AI is already paying attention to, then expand.
Third-Party Signals: Getting Mentioned Beyond Your Own Site
Your website is only one input. AI synthesizes answers from across the web, which means third-party mentions can matter just as much as your owned content. Building entity authority through citations is the process of moving from “getting a link” to “getting mentioned in the broader data ecosystem AI systems rely on.
✔️ Close your citation gaps first. A citation gap is when a trusted third-party source mentions your competitors but not you. These are your highest-priority targets. Rather than guessing which outlets to pitch, identify the top 20–50 domains already influencing AI outputs in your niche, and focus there.
✔️ Pitch citable data. “How [Your Brand] reduced CRM data entry time by 40% for 500+ sales teams” is something an AI can extract and cite. “[Your Brand] is an innovative solution” is not.
✔️ Build reviews with substance. On G2 and Capterra, encourage customers to leave detailed reviews that mention use cases, integrations, and specific outcomes – not just star ratings. These anchor your brand as the consensus choice in AI training data.
✔️ Show up in niche communities. Reddit, Hacker News, and industry Slack communities are heavily indexed by AI. Authentic, helpful participation in relevant threads builds the kind of social proof AI treats as genuine signal.
Technical Readiness for AI Bots
Many companies are accidentally invisible to AI crawlers – not because their content is bad, but because their technical setup blocks the bots.
✔️ Check your robots.txt. Make sure GPTBot (OpenAI), ClaudeBot (Anthropic), and PerplexityBot are not blocked. Some security plugins block unrecognized bots by default, meaning AI models simply never see your content.
✔️ Add an llms.txt file. It’s underused and high-impact:
"Standard robots.txt is for indexing; llms.txt is for training. It gives AI a condensed factual summary of your core offerings – so it doesn't hallucinate your pricing based on old blog posts."
Lily Grozeva
Founder Saltanat Labs
Don’t bury your best content. AI crawlers are lazy – if your technical specs, compliance guides, or pricing are more than two or three clicks from your homepage, they’re unlikely to be surfaced. Critical content should be reachable in 1–2 clicks.
How to Measure Whether It's Working
Traditional metrics like rankings and traffic don’t tell the full story anymore. In AI search, success is measured by your brand’s presence within the synthesized answers users see – often before they ever click a link.
Two metrics matter most:
- AI Share of Voice (SOV) – how often your brand appears in AI-generated answers for your targeted prompts. This is the central metric of 2026.
- Citation quality – not just how often you’re mentioned, but how you’re framed. Being described as a “market leader” is very different from being mentioned as a “controversial option.”
Tools to track this: RankScale, Peec, Scrunch, Profound, and Knowatoa. Run a baseline report across your 20+ targeted prompts, then track monthly. Look for which AI models cite you, on which prompts, for which pages, and alongside which competitors.
The 70/30 Rule
A simple way to think about effort allocation: 70% traditional SEO, 30% AI-specific work.
👉 Strong SEO – good rankings, authoritative content, a well-structured site – is still the core engine of AI search success. Google and Bing results are actively used as sources by AI models, so traditional rankings still matter.
👉 The 30% AI-specific layer covers prompt research, schema markup, llms.txt, ungating content, Bing optimization, and third-party citation building. These compound on top of solid SEO – not instead of it.
And the most important reality check: AI visibility is a growth multiplier, not a shortcut. For startups still finding product-market fit, that comes first. Once you have something genuinely worth recommending – AI search is a powerful way to make sure the right people find it.
Where to Go From Here
AI visibility isn’t a one-time project. It’s an ongoing practice. The search landscape is shifting fast, and the companies that stay discoverable will be the ones that treat this as a continuous process: researching prompts, refreshing content, building citations, and tracking what’s working.
If you missed Part 1, which covers how AI search works, why traditional SEO is no longer enough, and the first practical fixes , you can read it here.
And if you want the full framework in one place, the AI Visibility Founders Playbook goes deeper on every step across both parts, with practical frameworks, tools, and checklists you can act on right away. Download it here.
