A landing page the campaign and the agent can both use

The page has three jobs: match the line the person clicked, answer the questions the agent will be asked next, and fire one inquiry event. If the ad, the form, and the agent each invent their own story, Monday’s number is three numbers. Build one surface both systems can stand on.

The ad said the kit ships in 48 hours.

The page opened on a brand film and a paragraph about craft. The form asked for a company size. The widget on the corner offered to “help with anything.” Two people wrote in. Four more started a chat and vanished. The account was “working.” The inbox was not.

That is a landing-page problem, not a media problem.

Google’s own instruction is older than the agent: keep messaging consistent from ad to landing page, and make the page follow through on the offer. It sits in Quality Score guidance and again in Optimize your ads and landing pages. The agent does not get a waiver. If a person can arrive from Meta with one sentence in their head, the page has to say that sentence back — and leave a structured answer the agent can hand to the desk.

We design that surface under marketing and web. The weekly board for Meridian is ads, spend, and the pages. Not a PDF of reach.

Why do the campaign and the agent need the same page?

Split them and you pay twice.

The campaign needs a URL that matches the promise, loads on a phone, and converts the click it already bought. The agent needs facts it can cite: what is included, what is not, how long, how to buy, when a person must take over. Those are the same facts. Writing them twice is how they drift.

G2’s 2026 research found 51% of B2B software buyers now start research in an AI chatbot, up from 29% — The Answer Economy. Some still land from an ad. Some arrive half-briefed by a model that quoted a competitor. A page built only for a cold click fails the second visitor.

Unbounce’s Q4 2024 Conversion Benchmark Report — 41,000 landing pages, 464 million visits, 57 million conversions — puts the median landing-page conversion rate at 6.6% across industries (Unbounce, March 2025). That is a median, not a target. It is useful as a reminder that most pages already leak. Adding a second, contradictory conversation on top of a leaky page does not make a system. It makes two leaks.

One URL. One offer. One inquiry. The ad points at it. The agent reads it. The form and the handoff write the same event.

What does message match actually mean on the page?

Message match is not stuffing the keyword into the H1.

Google’s example is concrete: if the keyword is discount shoes and the ad promises 20% off, the page has to sell shoes at that price. If the ad asks someone to sign up for a free tour, the form has to be the tour. Their landing-page note calls the ad-to-page connection the bridge between a person and a purchase. Weak bridges bounce.

On Meta the click is a thumb-stop, not a query. The person saw a line and a picture. The first screen of the page has to be that line and a related picture. If the creative said “the board your Monday meeting can run,” the page may not open on a generic “we are a full-service studio.” We have a whole note on the homepage test. A landing page is stricter. Four seconds, and the stranger must recognise the ad they just left.

Practices we hold:

• The eyebrow is the campaign name or the exact offer, not the brand manifesto.
• The H1 is the ad’s promise in the page’s type. Same verb.
• The first sentence is proof or constraint — ships in 48 hours, one weekly board, nights covered — not a mission.
• The CTA uses the same verb as the ad. “Book the audit” does not become “Submit.”
• The image is a continuation of the creative, not a new photoshoot the ad never showed.

If you cannot keep the landing page still while you test creative, you will never know whether the line worked. That is the method in the ad creative that survived.

What structured answers does the agent need to hand off?

An agent that takes action cannot improvise the commercial facts.

A chatbot answers. An agent takes action. The difference shows up in the inbox — we wrote that out in chatbot vs AI agent. For the agent to file an inquiry instead of chatting, the page has to publish the answers in a shape a model can retrieve without inventing.

We put those answers on the page as a visible FAQ and as copy a person would actually read. Hidden JSON-only blocks that contradict the prose are how you get two truths. The person reads the prose. The agent reads both. They must match.

The block we keep seeing work, as a checklist, not a template:

Princeton’s GEO paper (Aggarwal et al., KDD 2024 / arXiv 2311.09735) found that citations, statistics, and quotations can lift visibility in generative engines by up to about 40% in their tests. We do not treat that as a traffic forecast. We treat it as a reason to put sourced facts on the page the agent reads — named, dated, linked — instead of adjectives.

If the offer changes, the FAQ changes the same day as the ad. Stale structured answers are worse than none. The agent will cite them with confidence.

What is the one inquiry event?

If the form writes generate_lead and the widget writes ChatStarted and the phone link writes nothing, you do not have a number.

The event is one: a form, a booked slot, or a call that lasted. How we sit with that number is the Monday note. This page is only where the event is born. The agent handoff is the same action with a source field.

Google Ads will happily optimise for the event you give it. Give it page_view dressed as a conversion and the account will buy you page views. Meta will do the same with a pixel that fires on scroll. The landing page is where you refuse that.

Implementation, as steps, because this part is a sequence:

1. Name the event inquiry in the tag manager, the ads accounts, and the agent.
2. Require an email or a phone. A nameless “booked” is not an inquiry.
3. Pass source, campaign, creative_id, and the offer slug.
4. Deduplicate. Form plus agent on the same email in ten minutes is one inquiry.
5. Write the row the Monday board reads. If it only lives in Ads Manager, it will be argued.

On Meridian the weekly room is one board: the ads, the spend, the pages. The inquiry is the shared KPI. Creative that cannot produce it gets killed. Pages that cannot accept it get rewritten. Not both in the same week if you can help it.

What does the annotated structure look like?

Top to bottom, on a phone first. Desktop can have more air. It cannot have a second story.

1. Eyebrow

Campaign or offer. Eight words or fewer. This is how the visitor confirms they are in the right place. If the ad was “Weekly board for Meta and Google,” the eyebrow is that, not the company name.

2. H1

The promise. One line. The type can be beautiful. The sentence cannot be vague. If the homepage is allowed a mood, this page is not.

3. Proof, immediately

A constraint or a fact. “One board. The ads, the spend, the pages.” Or a named result you are allowed to print. Not a carousel of logos that have nothing to do with the offer.

4. The offer block

What they get, what they do, how long. The CTA repeats the ad verb. Secondary link is allowed if it is the same offer in a calmer form — “see a board” — not a tour of the whole studio.

5. The form

Only the fields the desk will use this week. Name, work email, the question the ad implied. Company size is a later conversation. Every extra field is a reason to open the agent instead and then abandon it.

The same fields are the schema the agent collects. If the form asks for use_case and the agent asks “how can I help?”, you will not be able to join the rows.

6. A short stretch of explanation

One or two blocks. How the week runs. What you will not do. This is for the person who scrolled and for the agent that needs a paragraph to quote. It is not a blog post.

7. FAQ, visible

Five to eight questions. Written as questions people type. First sentence is the answer. This is also the agent’s brief. If you would not let the agent say it, do not publish it.

8. What happens next

Time and person. “We reply within a day with three lines: what we understood, what we would measure, what we would turn off.” If you cannot keep that, do not write it.

9. Footer that does not restart the site

Legal, a text link home, nothing that looks like a second campaign.

Unbounce calls 10% or above “good” in general, and says the real bar is industry and offer (same note). We do not chase 10%. We chase a page whose inquiry the desk can sit with.

How should the page talk to the agent without becoming a widget farm?

The agent is not the hero of the landing page.

If the first thing the paid click sees is a bubble, you have told them the offer is a conversation. Sometimes that is true — after-hours, a configurator, a qualification the form cannot hold. Usually it is an apology for a page that would not decide.

Rules we use:

• The agent is available. It is not the headline.
• It can fill the same fields as the form and submit inquiry.
• It cannot invent a discount, a lead time, or a feature the FAQ does not state.
• It must hand off when the visitor asks for a price you do not publish, a legal commitment, or anything marked “person only.”
• The transcript is stored. The Monday number is still the inquiry, not the transcript count.

YourGPT shows up here because that is the product we put on client sites. The knowledge is their writing — help centre, offer page, the FAQ above. The landing page is part of that knowledge. If the agent is trained on a generic site and the paid URL says something sharper, the agent will sand the offer back down to generic. Train it on the page you are buying clicks for.

Aurelia’s desk is the support version of this: the agent answers from their help centre and files the ticket. The marketing version is the same shape. Different knowledge. Same refusal to let the widget become a second brand.

What should you measure in the first two weeks?

Not ROAS. Not a story about view-through.

The first two weeks of a new page are for three numbers:

Google suggests using conversion rate as a proxy for landing-page experience, while being clear it does not change Quality Score status (Quality Score article). We agree with the proxy and ignore the status-chasing. Fast and mobile still matter.

If cost per inquiry is stable and the desk says the conversations are real, leave the page. Test the picture and the line. If the conversations are wrong, the page is lying — usually in the H1 or the FAQ. Fix the lie.

Meridian’s rebuild was this at account level: one board, the ads, the spend, the pages. The landing page is a column on that board, not a side project.

What do we refuse to put on this page?

A second offer. A blog teaser. A full navigation. A calendar that books a 45-minute tour for a $20 question. A chatbot greeting that says “Hi! 👋 How can I help you today?” as if the ad had not just said the thing.

Also: a different promise per ad group all pointing at this URL. Message match dies when five headlines share one H1. Either tighten the campaigns or make a second page.

And we refuse to call a scroll-depth event an inquiry. A long chat feels like work. Work is not demand.

FAQ

Should a landing page for ads also host the agent?

Yes, if the agent submits the same inquiry event and cites the same offer. No, if the agent is a generic helper that restarts the conversation. The page stays the source of truth. The agent is a door into that truth. A bubble that cannot file a record is decoration on a paid URL.

Do I need a separate landing page per campaign?

When the promise changes, yes. When only the picture changes, no. Google wants the page to match the ad and the keywords. Meta will send you traffic that saw one line. If two campaigns cannot share an H1 without lying, they cannot share a URL. Creative tests can share a URL. Offers cannot.

What is a good landing-page conversion rate in 2026?

Unbounce’s Q4 2024 median is 6.6% across 41,000 pages; they call 10% “good” in general and warn that industry and offer move the bar. Use the median as context, not a KPI. An inquiry the desk can sit with is the rate that matters. A high rate of the wrong event is how accounts look healthy until Monday.

How do I keep the agent from contradicting the ad?

Publish the commercial facts on the page and retrieve from that page. Ban unpublished prices, lead times, and “we can probably do that.” Review transcripts against the FAQ weekly. If the agent says something the page does not, that is a knowledge bug, not a model bug. Fix the source.

Should the form be short if the agent can collect the rest?

Yes. The form should be the fields you need to reply. The agent can gather context after the inquiry exists. Asking the form to qualify like a salesperson is how paid traffic disappears. Asking the agent to qualify without ever writing inquiry is how you get interesting chats and an empty board.

Does message match still matter with AI Max and Advantage+?

More, not less. Google’s AI Max can expand final URLs toward pages it predicts will perform. Meta’s ranking will pick among the creatives you give it. If those systems can choose, the destinations still have to keep the promise. A model cannot rescue a page that changed the offer.

What should I send a studio if I want this page built?

The live ads (or the three you would run), the current URL, a week of inquiries, and the questions the desk is tired of answering. We will write back with the H1 we would hold, the FAQ the agent would be allowed to cite, and the single event we would fire. Use the contact form. Do not send a sitemap.

The page is finished when a stranger who clicked the ad recognises the offer, a stranger who asked the agent gets the same offer, and Monday can count both as one number. Everything else is furniture.