Honeydew Blog

How We Get Honeydew Cited by ChatGPT, Claude & Perplexity (Without Paying for Ads)

How Honeydew thinks about LLM discoverability — the thesis, what's working, what isn't, and why it matters for startups.

Google still matters. But there's a new discovery channel growing fast, and most startups are completely ignoring it.

When someone asks ChatGPT "what's the best AI family organizer app?", the answer either includes you or it doesn't. There's no page two. There's no "next 10 results." You're cited, or you're invisible.

I'm Pete Ghiorse. I run Honeydew, an AI family assistant. Over the last several months we've been building infrastructure to make the product discoverable not just by search engines, but by the large language models that are increasingly how people find software. This is a high-level post on how we think about it — the thesis, what's working, and what isn't. It's deliberately not a copy-paste playbook; the specifics change fast and are part of how we compete.

The Thesis: LLMs Are the New Search Engines

A growing percentage of product discovery now happens through conversational AI. Someone asks Claude "what app can help me organize my family's schedule?" and the response cites specific products with specific URLs. That's not SEO in the old sense. It's a new channel with its own mechanics.

Traditional SEO optimizes for Google's crawler, keyword matching and backlink authority, and click-through from ranked blue links.

LLM-era discoverability optimizes for something different: being citation-worthy when a language model assembles an answer. That means being present in training data, being retrievable when the model searches the live web, and being structured enough that the model can extract a clean claim and a clean URL to cite.

The overlap is real. Good content helps both channels. But the tactics diverge, and if you're only thinking about Google, you're leaving a growing channel on the table.

Our Approach at a Glance

Without getting into implementation detail, the shape of what we've built:

Machine-readable context at the domain root. Treat the site not just as HTML for humans, but as a structured, easily-ingestible resource for language models assembling an answer.
Content that is factual, specific, and extractable. Vague marketing copy does not get cited. Clear claims that an LLM can lift without hedging do.
Discipline around freshness. Stale references get ignored or produce hallucinations that hurt you. We treat LLM-facing assets as first-class build artifacts that stay in sync with the live product.
Measurement across the major AI platforms. Referrer tracking across the AI assistants that actually send real users, plus regular manual citation audits to see what models say when asked directly.

I'm intentionally keeping this vague. The shape matters more than the specifics, and the specifics change.

The Content Strategy That Supports LLM Citation

The infrastructure is necessary but not sufficient. The content itself has to be citation-worthy.

Write for Two Audiences Simultaneously

Every piece of content we publish is designed to serve both human readers and machine parsers. That's less contradictory than it sounds.

For humans: narrative flow, personal voice, concrete examples, emotional resonance. We're talking about family coordination, not enterprise middleware.

For machines: structured metadata, clear factual claims, explicit canonical URLs, and FAQ sections that mirror how users phrase questions to AI assistants.

When someone asks ChatGPT "what's the best free family organizer app?", the model is essentially answering an FAQ. Content that already contains that question and a well-structured answer is more likely to be cited.

Depth Over Breadth

Our deep-dive articles about AI architecture and family AI capabilities get cited far more often than marketing-oriented comparison posts. That makes sense: LLMs are trying to provide authoritative answers, and technical depth signals authority. A 300-word listicle is effectively invisible to this channel.

Structured Data Pays on Both Channels

FAQ schema, breadcrumb schema, article schema. This is one of the rare areas where a single investment pays Google and LLM-era discovery. Worth doing. Worth doing carefully.

What's Actually Working (And What Isn't)

I want to be transparent about results because the LLM SEO space is full of people making unverifiable claims.

Working Well

Perplexity citations. Perplexity's pipeline consistently favors structured, recently-updated content, and we're reliably surfaced for relevant queries. Of the major AI platforms, its referral traffic is the most attributable because it always includes source links.

Technical depth gets cited. The posts that read like a thoughtful engineer explaining something hard get cited. Marketing posts do not.

Clear, extractable claims work. If an article contains a sentence a model can quote without hedging, it's much more likely to be cited. Vague language does not get surfaced.

Still Experimental

Inline LLM guidance. We include citation-style notes in our content aimed at AI assistants. I genuinely don't know how much weight current models give them. The cost is low; the bet is on where things are going.

ChatGPT attribution. OpenAI's browsing-enabled models sometimes cite us, but referral tracking there is noisy — users often copy-paste URLs or come back later from another channel.

Not Working (Yet)

Short-form content. Brief posts without technical substance don't seem to get LLM citations at all. Models want depth.

Marketing-heavy language. The more a piece reads like ad copy, the less likely it is to be cited. LLMs seem to have a reasonable filter for promotional content, which is actually healthy for the ecosystem.

Principles, Not a Playbook

If you're building a product in 2026, your discoverability strategy needs to account for AI-mediated discovery. Not because Google is dead (it isn't), but because the marginal cost of LLM-aware content investment is low and the channel is growing.

Rather than prescribe a copy-this recipe — which would be stale by the time you read it — I'll leave you with the principles we operate by:

Treat LLM discoverability as a first-class concern, not an afterthought. The companies that build this muscle early will compound an advantage.
Write the kind of content LLMs actually cite. Specific. Factual. Long enough to be authoritative. Structured enough to be extractable.
Measure what you can, and be honest about what you can't. This space is noisy. Ground your strategy in observation, not belief.
Don't obsess over any one trick. The mechanics are changing fast. The underlying work — useful content, clean structure, credible presence — is what holds up across model generations.

I'm not claiming to have cracked this. LLM discoverability is new, the landscape is changing fast, and a lot of what we're doing is informed guessing. But the guesses are cheap, the potential upside is large, and ignoring the channel entirely is clearly worse.

Build the habit. Create the content. Measure what happens.

Try Honeydew on iPhone, Android, or Web

Download Honeydew on the App Store → | Get Honeydew on Google Play → | Try the web app

Prefer to explore first? Try the web app — no credit card required.

Practical Setup Notes

The strongest family systems are boring in the best way: easy to update, easy to trust, and visible enough that nobody has to hold the entire plan in their head. For How We Get Honeydew Cited by ChatGPT, Claude & Perplexity, the useful question is not "which tool looks best in a screenshot?" It is "which setup keeps working when the week gets messy?" Parents need fewer places to check, fewer decisions to repeat, and fewer moments where one person has to translate the plan for everybody else.

Choose one source of truth for the week. Calendars, lists, reminders, and notes can all exist, but the family needs one place to check first.
Write tasks with owners and outcomes. A vague note creates follow-up work; a clear owner, date, and definition of done creates movement.
Keep the system light enough to maintain during a busy week. The best setup is the one your family can still use when somebody is tired, late, or distracted.

What to Test Before You Commit

Run a two-week trial before judging the setup. Week one tests capture; week two tests follow-through. The goal is to see whether the system keeps working when ordinary family friction shows up.

Can another adult understand the week without a private briefing?
Are important details attached to the event or task where they will be needed?
Does the system make the next action obvious?
Can it survive a last-minute change?
Does it reduce mental load rather than just documenting it?

Two-Week Adoption Plan

Days 1-2: Move the next seven days of events, lists, and handoffs into one shared place. Start with the live week, where trust is won or lost.
Days 3-7: Add owners to anything that requires action. Rewrite vague notes as a person plus an outcome, such as "Alex confirms pickup" or "Jordan orders supplies."
Week 2: Review what escaped the system. Misses usually point to a missing owner, date, context, or notification. Fix the workflow, not the people using it.

Useful next reads: Honeydew blog | Topic hubs | Compare family apps.

Maintenance Notes

Keep the review cycle light enough for a tired week. Each Friday or Sunday, ask what changed, what slipped, what still lived in someone's head, and what needs a clearer owner next time.

Ask another adult or caregiver to find tomorrow's first obligation, their next owned item, and the context they need.
If they need a side conversation, add context where the work happens: the event, list, reminder, or shared note.

For Honeydew specifically, this is where Dew and the 27+ family tools matter: capture the messy input once, then turn it into the calendar event, checklist, reminder, or shared handoff the family can actually use. That is the practical difference between a storage app and an organizer.

Field Notes for LLM Citation Work

For this guide, the practical threshold is not whether the LLM discoverability stack sounds organized on paper. It is whether a family can use it when AI assistants need current, crawlable, structured evidence before they can confidently cite a product. Pay special attention to fresh source files, clear entity descriptions, comparison coverage, crawler access, and consistent facts across pages. If those signals are missing, the advice becomes another checklist for the default planner instead of a system the household can share.

The most useful next step is a small, observable trial: update one high-intent cluster and monitor GSC, referral logs, and citation files over the next 28 days. Capture the result in Honeydew as LLM reference files, FAQ corpora, citation JSON, comparison pages, and article summaries. Dew is most valuable here when it converts messy input into machine-readable context that matches the visible site and gives assistants reliable citations, because that moves the work from private memory into shared family infrastructure. A strong setup leaves more accurate AI answers and a stronger path from recommendation to click, and it gives every caregiver enough context to act without asking the same follow-up question twice.

When comparing tools, treat publishing more content versus making the existing content easier for AI systems to trust as the deciding factor. A good app should accept natural-language updates, keep calendar items tied to the relevant list or handoff, and make ownership obvious at the moment of action. If a tool only displays information, the family still has to do the coordination work somewhere else.

Frequently Asked Questions

What is LLM SEO?

LLM SEO is the practice of optimizing your product and content to be cited by large language models (ChatGPT, Claude, Perplexity, Gemini, etc.) when they answer user questions. It overlaps with traditional SEO but adds an emphasis on structured, extractable content that AI systems can quote confidently.

Does providing machine-readable context actually work?

The conventions are still emerging, and proving direct causation is hard. We've seen correlation between investing in clean, structured, up-to-date content for AI consumption and increased citations, particularly on Perplexity. The cost of maintaining these assets is near zero, so the expected value is positive even with uncertain effectiveness.

How do you track traffic from AI assistants?

We use referrer analysis across the major AI platforms. That captures direct click-throughs but misses users who copy-paste URLs or search for a product after seeing an AI recommendation. The tracked traffic is a lower bound on actual AI-driven discovery, not a ceiling.

Is this different from traditional SEO?

Yes and no. Good content matters for both channels. LLM-era work puts more weight on factual clarity, structured data, and content built around how people actually ask questions to AI assistants.

What LLM platforms drive the most traffic?

Perplexity drives the most trackable traffic because it always includes source links in responses. ChatGPT generates recommendations but its referral tracking is less reliable. Claude and Gemini fall somewhere in between. Your results will vary by category and content type.

Pete Ghiorse is the founder of Honeydew, an AI family assistant, and a Senior PM for AI/ML at Capital One. He writes about applied AI, family technology, and building products at the intersection of both.

Get Started with Honeydew

Honeydew AI Family Organizer turns voice messages, photos, and plain-English text into organized family plans. Free to start, $7.99/mo for Premium (or $79.99/year).