I built an AI assistant for content heavy sites and I do not run a content site myself. Honest feedback wanted.
I want to share what I am building and get reality checks from people who actually run content sites, because I do not, and I am aware that is a problem.
The product is an embedded AI assistant for content heavy websites. Visitors ask a question in natural language and get an answer pulled from the site's existing articles. Not a generic chatbot. It indexes the actual content and points visitors to actual pages.
The theory is that a lot of content sites have years of valuable articles that visitors never find. Search is weak, categories are broad, old posts get buried. Someone lands looking for "best 7 day Italy itinerary" and even if you have the perfect article for that, they may never see it. They bounce. You lose the pageview, the affiliate click, the newsletter signup.
So the product tries to fix that. Visitor asks a question, assistant finds the relevant articles, returns a structured answer with source links and recommended reading.
I built a demo travel site with 25 articles to test against, because I do not own a site and do not know anyone who does. It is focused on Paris specifically, which turned out to produce better answers than going broad. Focused archives compound in retrieval quality.
Here is the landing page with a short video demo: https://asksite.carrd.co
Where I am uncertain:
-
I do not know if this is a real problem or a nice to have. The theoretical value is clear. Whether site owners actually care enough to install a new tool is a different question.
-
I do not know if the chat or the analytics layer is the bigger value. Every question a visitor asks is a signal. What people ask, where content gaps are, what converts. I suspect the analytics may be the real product. But that is a hypothesis.
-
I do not know what the right ICP is. My current thinking is travel and points, recipes, and DIY tutorials. Less sure about restaurant guides, local food blogs, or news adjacent sites.
If you run a content heavy site, or you have built products for people who do, I would genuinely value:
-
Does the problem resonate, or does it sound like a solution looking for a problem?
-
What would make you actually install something like this on your site, versus politely nod and move on?
-
Is the analytics part interesting to you, or is that my own bias showing?
Happy to share more about what I am building if useful. Not pitching, just trying to get real feedback before I spend another month on something that might not matter.
The problem definitely exists .. but it feels uneven depending on the site.
For something like travel or tutorials where people arrive with specific questions, this makes a lot of sense. For more casual browsing sites, it might just get ignored. The analytics angle is interesting though. Seeing what people tried to find but couldn’t is probably more valuable than the answers themselves. The challenge is whether site owners trust it enough to add another layer to their site, especially if it affects speed or UX.
What kind of feedback have you had so far from people actually running content sites?
Thanks for engaging with this seriously.
The "uneven depending on the site" point is right. I have been thinking about this same thing in terms of intent. Sites where visitors arrive with specific questions (travel, recipes, tutorials, how-to) probably get real value from this. Sites where visitors arrive to browse or kill time (entertainment, news, lifestyle aggregators) probably do not, because the assistant assumes intent that the visitor does not have.
Your "what people tried to find but couldn't" framing is sharper than how I had been thinking about the analytics. I was framing it as "see what visitors ask" which is just observation. Your version turns it into a content gap signal, which is a different and more useful thing for a site owner. Will steal that framing if that is okay.
On the trust question. This is the part I am wrestling with most. Even if the product worked perfectly, asking a site owner to add another script tag, another widget, another thing that could affect page speed and UX, is a real ask. I do not have a clean answer for this yet beyond starting with people who already trust me, which I do not have, or starting with people who are willing to take a risk, which is a small group.
On feedback from real site owners, honest answer is not much yet. This post is part of the validation process. I have one founder running an adjacent travel product who has been giving me real feedback on positioning and ICP, and I am hoping posts like this surface a few more. So far you and a couple others. No actual content site owners have replied yet, which is its own data point.
What pulled you into this post? Are you in this space yourself, or close to it?
That makes sense, especially on the trust side. Adding anything that touches speed or UX is always a harder sell than the product itself.
What pulled me in was the intent angle. A lot of tools assume users are looking for something specific but in reality a big chunk are just browsing with no clear goal.
I’m not in this space directly, but I work around automation and user behaviour quite a bit, so it’s interesting seeing how that plays out on content sites.
The intent angle is the thing I keep coming back to. Most of the tooling assumes intent is uniform when it is anything but. Another founder on IH I have been talking with made a related point that has been rattling around. He said the harder problem is not retrieval quality but whether the widget earns the next click in the first 5 seconds. That only matters if you have already correctly assumed the visitor has intent in the first place.
If you have a moment, I am genuinely curious about something from your automation and user behavior background. When you have seen tools introduced into existing user flows, what tends to determine whether users adopt the tool versus ignore it? I have been treating this as a positioning and visibility problem (where the widget appears, what it looks like, what the first message says). I suspect there is more to it than that, and your perspective might surface something I am missing.
No pressure, just appreciate the lens.
From what I’ve seen, adoption usually comes down to timing more than anything. If something shows up while the user is already trying to solve a problem, it gets used. If it shows up while they’re just browsing . . it gets ignored no matter how well it’s positioned.
The other part is how obvious the value is in the first few seconds. If it feels like it will save time immediately, people try it. If they have to think about it, they skip it.
So less about the tool itself . .more about catching the right moment with a clear payoff.
This is really useful, especially the timing part. The implication for us is that the widget needs to surface when the visitor is already mid-problem, not as a passive button waiting to be clicked. On a travel site that probably means showing up after they've scrolled through a guide and slowed down, not the moment the page loads. On a recipe site it might be after they hit a step that needs substituting.
The "obvious value in seconds" part is the harder one. The chat icon itself doesn't do any of that work. The first message it shows might need to. Something like an actual question pre-filled based on the page they're on, not a generic "ask me anything." Curious if that matches what you've seen work.
Indeed . . that matches.
The pre-filled question idea is strong. It removes the “what do I even ask?” friction and turns it into a continuation of what they’re already doing.
I’ve seen similar where the biggest drop-off isn’t the tool . . it’s that first step. If that’s not obvious, people don’t engage at all.
Feels like the trigger plus first message combined is what makes or breaks it.
Indeed, that's the right framing. "Trigger plus first message" is what we've been circling without quite naming.
We've been treating the chat icon as the entry point but the first message is what actually does the entry work, that distinction is going to change how we think about the whole onboarding moment.
I know a few content-heavy website owners personally who might be willing to answer your questions about this for free. Let me know if you want me to pass them along.
Yes please, that would be genuinely useful.
Happy to make it easy on your end. If it helps, you can either send them a one-liner pointing them at the IH post, or pass me their contact and I'll reach out directly with whatever framing works best for the relationship.
Either way, thank you.
The analytics hypothesis is probably right, and I'd bet on it over the assistant UI as the primary value.
Site owners don't install tools for visitors, they install tools for themselves. The chat interface is what visitors see, but what makes an owner care enough to add another script to their site is what they get out of it. "Here are the 50 questions your visitors asked this month that you have no article for" is a dashboard someone opens every week. A floating chat widget is something they set up and forget.
The ICP question is the harder one. Travel and recipes make sense on paper because the archives are deep and the content is evergreen. But the person who runs a 200-article Italy travel blog is usually a solo creator who's already overwhelmed. They may love the idea and never install it. The person who runs a mid-size recipe site with a small team and actual revenue goals has both the motivation and the bandwidth to try new tools.
One question worth testing: do site owners already know their search is broken, or do you have to convince them first? If they know, you're selling a solution. If they don't, you're selling a problem, much harder conversation.
The analytics-vs-widget framing is exactly where we've landed. A widget gets set up and forgotten. A dashboard with the right questions on it gets opened weekly. That's the asymmetry.
The size and team structure point is sharper than the vertical cut. A 200-article solo creator might love the idea and never install it. A mid-size operator with a small team and revenue goals has both motivation and bandwidth. We're starting in affiliate-driven travel for credibility reasons, but within that, you're right that the operator profile matters more than the vertical does. Solo hobbyists are the wrong target inside any vertical.
On the search-broken question, honest answer is most don't know. Site owners look at GA4 and see traffic and bounce rate, not "what visitors couldn't find." Which means the first conversation is selling a problem, then a solution. The bet is that once they see the gap report, they can't unsee it. The challenge is getting them to look in the first place.
The 'can't unsee it' bet is the right one — once someone sees the gap report they'll wonder how they ran the site without it. The challenge of selling a problem first is real but it sounds like you already have the wedge: show them the gap report before asking for anything.
On the search-broken question — I wonder if the entry point isn't site owners at all initially. Affiliate travel is interesting because those operators are unusually metrics-driven. They already live in GA4 and care deeply about every bounce. If anyone is going to notice 'visitors couldn't find what they came for' it's someone watching affiliate click rates obsessively.
What does your current acquisition plan look like — are you going direct to site owners or thinking about a different entry point?
The affiliate travel operator profile you described is exactly who we're starting with for that reason. People who already obsess over bounce rates and click attribution will see the gap report and immediately translate it into revenue. That's the conversation we want to be in.
Honest answer on acquisition, the plan is more wedge than plan right now. Warm intros via LinkedIn into affiliate travel operators I have one degree of separation from, a small handful of direct outreach to sites that match the profile, and the IH thread itself as a way to surface people like you who'd test it. Not a scaled motion yet, deliberately. The point of the first batch is to find out whether the "can't unsee it" moment actually lands, before figuring out how to repeat it at scale.
Your "entry point isn't site owners at all initially" line is the part I want to push on. What were you thinking? Affiliate networks themselves, agencies running these sites, content tooling platforms? Genuinely curious where you were going with that.
The wedge-first approach makes sense, finding the 'can't unsee it' moment with 5-10 operators before systematizing is exactly right. Trying to scale before you know what lands is just burning resources.
On the entry point question, the thinking was that affiliate travel operators, even the metrics-obsessed ones, are still one person running one site. The ceiling on that motion is low if you're going one by one.
The more interesting entry point might be the affiliate networks themselves, ShareASale, CJ Affiliate, Impact. They have direct relationships with hundreds of operators simultaneously. If a network recommends your tool as something that improves site quality and therefore conversion, you get distribution without doing individual outreach.
Second option is content tooling platforms, Mediavine - Raptive. These are ad management platforms that already sit inside affiliate travel sites. Their users are exactly your profile, mid-size operators, revenue-driven, already paying for tools that improve performance metrics. A partnership or integration there gets you in front of thousands of the right operators at once.
Both paths are partnership plays, not direct sales. Different motion entirely but potentially much faster than one-by-one outreach.
I’ve been exploring the idea of AI-driven content platforms, but I keep coming back to one question:
Are we building more content… or are we actually improving human connection?
Most AI content-heavy apps today are focused on generating, optimizing, and distributing content at scale. But that still leads to more scrolling, more passive consumption, and less real-world interaction.
What I’m working on takes a slightly different direction.
It’s a social platform concept where AI doesn’t just generate or recommend content—it helps create real-world experiences. Users are matched locally, and AI generates simple “challenges” or prompts that encourage them to meet up and do something in real life together before they can post anything.
So instead of AI fueling endless feeds, it fuels real interactions that then become the content.
The idea is to keep the content layer, but reverse the order:
experience first → content second.
I’m currently looking for a technical co-founder who’s interested in AI + consumer social products, especially anything that pushes beyond traditional content consumption into behavior change and real-world engagement.
Would love to connect with anyone building in the AI content or next-gen social space.
The analytics angle is the real product, knowing what your visitors are actually asking is more valuable than the assistant itself and it tells you what content to write next. The ICP Q answers itself from there, whoever is most obsessed with content gaps ans SEO is your buyer - which point more to niche affiliate sites than travel blogs.
Strong read. We've landed on the same conclusion on analytics. The bullet that keeps getting sharper for us is "what visitors tried to find but couldn't" rather than just "what they asked," because that's the content gap signal a site owner can actually act on.
On ICP, you're right that affiliate operators feel the content gap pain most acutely. We're starting inside that profile, specifically affiliate-driven travel sites, because they have the same content gap obsession but the vertical is one I can have credible conversations in fastest. Recipes and DIY affiliate are the natural next slices. Same buyer pain, different content.
"What they tried to find but couldn't" is the sharper frame - tha'ts the frustrated exit signal, not just curiosity. Makes sense to start affiliate travel where you can have credible conversations quickly. Curious whether you're seeing that gap signal alone is enough to drive upgrades, or whether site owners need to see a few wins first before they trust the data enough to act on it.
Honest answer, I don't know yet, but my working hypothesis is that the gap signal alone isn't enough. A site owner looking at "47 questions your articles don't answer" needs at least one of those to land as undeniably true before they trust the rest of the list. Otherwise it's just another dashboard telling them things.
What I think does the work is the first time they see a question on the list that they recognize as a real gap they'd been half-aware of but hadn't named. That moment of "oh, I've been meaning to write about that" is probably what converts. Whether that happens often enough and fast enough to drive an upgrade is the biggest unknown for us right now.
That " oh i've been meaning to write about that" moment is the hook - it works because it's not telling them something new, it's naming something thay already half-knew. That's usually the highest-trust form of insight. The question is whether you can engineer the expirence to surface that movment early enough, before they've decided whether the product is worth paying attention to.
"Naming something they already half-knew" is sharper than how I had it. That's the form the insight has to take, and you're right that the timing constraint is brutal. It can't happen on day 30, it has to happen in the first dashboard view or you've already lost them.
That probably means the first batch of questions surfaced needs to be hand-curated, not algorithmic, at least early on. The product can get smarter later. The first impression can't be left to chance.
This is amazing idea to built an AI assistant for content heavy sites because it saves lot of time of customers to read and find content on your site. Instead using AI it gives you perfect landing page content depends on your query.
Thanks for the read, appreciate the kind words.
Welcome
This is a really thoughtful approach — and I like that you’re questioning it early 👍
The problem does feel real (buried content, weak search), but I think the key is how it’s positioned.
Right now it can feel like a “nice to have”, unless it clearly ties to:
→ more pageviews
→ more affiliate clicks
→ better conversions
The analytics part you mentioned actually feels very interesting.
What people ask is strong intent data — that could be more valuable than the chat itself.
Also your Paris insight is solid — focused/niche sites probably get way better results than broad ones.
If I were a site owner, I’d install this if I could see:
→ clear impact (like session depth or clicks increasing)
Curious — have you tried this on a real site yet or only your demo so far?
Also, I’m running a small project (Tokyo Lore) where we test tools like this with a focused group of builders and see how they actually perform in real usage.
Since you’re still validating problem vs value, this could be a strong fit to figure out what really sticks.
Happy to share more if you’re interested 👍
This is actually a really interesting idea, especially the part about surfacing older content that users would never find.
I’m not a content site owner either, but as a user, I can see how this could reduce bounce if the answers are actually relevant and fast.
The analytics angle you mentioned might be bigger than the chat itself. Knowing what users are searching for but not finding could be really valuable for content strategy.
Curious — how accurate are the answers right now when someone asks something specific? Does it ever return weak or irrelevant results?
Also, I’m working on a marketplace platform and running into a different problem (users sign up but don’t take action), so I can relate to trying to validate if something is a real problem or just an idea.
Would be interested to see how this evolves.
On accuracy, honest answer is we do not have great data yet, but our impression from testing is that it is okay, not great. With the focused 25 article Paris demo, common questions land well because there is meaningful content to retrieve from. Things like "best 7 day itinerary" or "where should I stay" return decent answers because there is an article that actually addresses those questions.
Where it gets weaker is when someone asks something specific that is only tangentially covered. The retrieval finds the closest match but the answer can feel generic, or it pulls from one paragraph in a longer article and loses the surrounding context. We are still figuring out how to handle the gap between having an article on a topic versus having an article that actually answers the question someone asked. My partners handle most of the technical layer so I am speaking at a higher level here.
On your marketplace problem, that is a really common pattern. The "signed up but did not take action" gap is its own validation problem because the signup says they are interested but the lack of action says something else, and you do not know which signal is real until you talk to some of them. Same kind of thing I am wrestling with from a different angle.
Will keep posting updates as this evolves. Thanks again.