Customer Calls Should Fix Your Pages Before They Become Posts
Your front desk answers the same handful of questions every week. The advice going around says to record those calls, hand the transcripts to AI and publish a post for each one. A customer call should first fix a page you already have, and only rarely start a new post. A repeated question usually points to an existing service, pricing or FAQ page that failed to answer it. Below is a routing rule that sends each question to the page that owns it.
HubSpot's own blog now brings in about a tenth of its content leads
HubSpot is the company most owners picture when someone says inbound marketing. Its blog peaked above 20 million organic visits a month and then lost about 80% of that traffic, CMO Kipp Bodnar told Greg Jarboe in a 29 October 2026 Search Engine Journal interview. Revenue still grew nearly 20% in 2025. The traffic collapsed and the money kept coming, while the leads moved somewhere else.
Bodnar put a number on that somewhere else. About 90% of HubSpot's content leads now come from channels that barely existed in its mix five years ago. The blog accounts for about 10%. If the company behind the inbound playbook gets one content lead in ten from its blog, a new post for every sales call is a bet on a shrinking slice.
None of the top three pages on turning call transcripts into content tells you where a call question should go. The question goes first to the existing service, pricing or FAQ page that failed to answer it. CTM, the call-tracking company, points recurring call questions at FAQ and service pages. Google flags mass-produced AI content published without oversight. HubSpot's lead numbers above tell the same story.
Picture a physio clinic where three callers a week ask whether they need a GP referral before they can book. Every one of those calls ties up the receptionist. It also hints at the callers who never phoned and booked somewhere else. The referral answer belongs on the booking page and in the FAQ, in the words callers use: "Do I need a referral from my GP to book?" A separate blog post on referral rules puts the answer on a page the caller never opens.
I built and run my own live meeting transcriber, a tool whose speech-to-text runs on my own computer. I have also worked with medical clients, including a London ADHD clinic I booked solid for three straight months, where callers talk about their own health. Call data should fix the pages that answer callers, and it should never feed a tool that publishes what they said.
Calls-into-posts advice always ends at a drafting tool
Search for how to turn sales calls into content and the top results agree on what you should end up with. IMPACT's guide says to record every call and connect each transcript to your CRM, the system where you keep customer records. AI then finds the gap, drafts the page, and a human approves it. The guide recommends HubSpot's Breeze AI for the job. It also tells you to "multiply that by 150 calls a month," a pest-control figure IMPACT gives as an illustration, with no measurement behind it.
Kate Starr's Content Backstage workflow, published 5 April 2026 and updated 27 April, runs Granola on every call. A custom recipe then turns each call into questions, content ideas, and angles for posts, videos and articles. The piece is affiliate-led. Copy.ai's Kyle Coleman describes AI pulling insights from transcripts, scoring deal health and generating draft content such as blog posts.
All three assume the output is new content and that more of it is better. Each one points you toward a drafting tool: Breeze AI, Granola, or Copy.ai's own platform. None of them mentions what Google says about content produced at that volume with little review.
| Attribute | IMPACT auto-draft | Content Backstage idea-to-post | Fix the page first |
|---|---|---|---|
| What gets published | A new AI-drafted page for each gap found in a transcript | Post, video and article angles from every call | An edit to the service, pricing or FAQ page that missed the question |
| Human review point | Approval after AI has drafted the page | Not stated in the published workflow | Before anything is written, when a person picks the page |
| Scaled-content risk | High: output grows with call volume | High: each call yields several pieces | Low: edits to pages you already own |
| Privacy exposure | Every call recorded and connected to the CRM and an AI tool | Every call run through an AI note-taking app | Only the written question leaves the call, with no names |
| Which page gets better | A new page for each gap the AI finds | New posts, videos and articles | The page the caller should have found |
Dan Taylor reported in Search Engine Journal on 15 October 2026 that Google's AI-content guidance says to fact-check everything AI generates before publishing. Google's guidelines name mass-produced AI content without oversight as little-to-no-effort content. Gary Illyes said at Search Central Live that scaled content is becoming more of a problem than link spam. A pipeline that drafts a post from every call, approved in one click, matches what those warnings describe.
The cost shows up in what Google bothers to keep. I have written before about how publishing more blog posts can shrink the number of pages Google indexes. A post per call can speed that up. Meanwhile your service page, the one a caller would actually book from, still lacks the answer.
Buyers now type whole questions, and your existing pages have to answer them
A traditional search query runs 2 to 4 words. Prompts in AI tools now run 40 to 100 words, Bodnar told Search Engine Journal. Google's data points the same way. Shivani Mohan wrote on the Google blog on 19 May 2026 that the average US AI Mode query is about 3x the length of a traditional one. "Decide" queries, the "which of" and "which one" kind, are growing 40% faster than AI Mode overall.
A 60-word prompt is a phone call typed out. It carries the situation, the budget and the doubt your front desk hears every day. A buyer asking an AI tool which physio near them takes patients without a referral is making a decision. The page that answers that decision in plain words is the page with a real chance of being quoted.
| Where the question is asked | Typical length | What your page has to answer |
|---|---|---|
| Traditional search | 2 to 4 words | A topic, such as "physio referral" |
| AI tool prompts | 40 to 100 words | The caller's full situation and the decision they are trying to make |
| Google AI Mode (US average) | About 3x a traditional query | "Which one" choices, growing 40% faster than AI Mode overall |
Bodnar described the standard HubSpot now writes to.
"Our content has to answer the full question in plain, chunked language, the way a buyer would ask it."
Kipp Bodnar, HubSpot CMO, Search Engine Journal, 29 October 2026.
That sentence describes your call log. Every recurring question already arrives in the buyer's own wording. Put that wording on the page, then answer it fully in short paragraphs that make sense on their own. Bodnar also said HubSpot's visitors from AI answers convert at roughly 3x the rate of traditional search, and their deal cycles run about half as long. Visitors who convert at three times the rate and close in half the time are the reason to fix the page AI tools quote.
The CTM team reached the same place from the phone side. In a CTM webinar recapped by Heather Campbell for Search Engine Journal on 20 October 2026, Lisa Salvatore said recurring call questions can feed FAQ pages and the answers on the relevant service pages. Objections, she said, show where comparison or trust content is needed. She added that matching customer wording helps people find you but does not guarantee an AI recommendation.
Her colleague Brian Barranger analyzed 29 lead-call summaries. The recurring theme was prospects struggling to connect marketing spend to revenue, and he treated that limited set as an illustration, not proof. Twenty-nine calls is enough to spot a question worth answering on a page. It is too few to justify a content program.
Look at where Barranger puts the weight. The AI handles speed. The conversations supply the relevance. That relevance does the most good on the pages your customers already open.
Route each repeated question to the page that should have answered it
Start with the page closest to the money. When a caller asks what you do, how a service works or whether it suits them, the answer goes on that service page. Questions about cost, deposits, packages or insurance go on the pricing page. If you keep prices off the site, read why ChatGPT looks up your prices and what happens when your site hides them. Practical questions about parking, referrals, hours, cancellations or what to bring go in the FAQ.
A new post is the last branch, and it only applies when no existing page owns the question. A question about which treatment suits a desk worker with chronic neck pain could earn a post, because no booking or pricing page should carry that depth. Before anyone drafts it, run it through the three-part test for whether a new blog topic is worth writing. That post scores new ideas. This one routes the questions your existing pages failed to answer.
You need no software for any of this. A sheet of paper at reception with one column for the question and one for tally marks does the job. Ask whoever answers the phone to write each question the way the caller said it, then add a mark every time it repeats. If your team already keeps call notes in a CRM, use those. On Friday, the questions with the most marks go to the top of the page-fix list.
Bodnar has a test for whether the wording holds up.
"If a company could put its logo on your message and nothing would change, that message is adding to the noise."
Kipp Bodnar, HubSpot CMO, Search Engine Journal, 29 October 2026.
Few competitors could put their logo on your referral policy written in your callers' words. Any clinic in the country could put its logo on an AI post about what to expect at your first physio visit. The page fix passes Bodnar's test. The generic post fails it.
If an agency runs your site, ask them: "For last month's call questions, which existing pages did you update, and what did you change?" A list of new blog posts in reply means you are paying for the shrinking slice.
Sensitive calls stay out of tools that publish for you
Medical, legal and finance callers say things they would never post in public. Picture a workflow that records every call, sends it to a cloud AI tool and drafts web copy from it. Those words are now one approval click from your website. Before you connect call recordings to any AI tool, check your privacy obligations with whoever handles compliance for your business. Then read the tool's terms on storage and training.
My own transcriber turns speech into text on my own computer, and the audio stays on that machine. For a clinic, the routing rule needs only the written question with every name removed. That stripped question is the only part of a sensitive call that should ever reach a web page.
A fixed page shows up as a question your phone stops asking
Bodnar admits the measurement is messy, even at HubSpot's size.
"Attribution is imperfect." And: "The signal is correlative, and it's strong enough that we treat AEO as a core part of how we market."
Kipp Bodnar, HubSpot CMO, Search Engine Journal, 29 October 2026.
AEO, answer engine optimization, means getting your business quoted inside AI answers. Your version of Bodnar's signal costs less to collect. Keep the tally sheet running after each page update. If the referral question falls from three calls a week to one, the booking page is doing the receptionist's job. Give each update a few weeks of calls before you judge it.
Then ask your team to make a note whenever a caller says "I read on your site" or "ChatGPT told me." Add Google's AI answers if your callers bring them up. A few of those mentions a month tell you which pages people read and which sources your callers trust. Attribution software never hears them.
Some numbers feel like progress and move no money. Posts published per month is the most tempting. Transcripts processed and AI ideas generated come next. A calls-to-content pipeline pushes all of them up fast. HubSpot's own figures show traffic and revenue moving in opposite directions, with a drop of about 80% next to nearly 20% revenue growth. Track repeat questions, phone mentions and bookings from fixed pages, and leave output counts off the report.
Frequently Asked Questions
Should I turn my sales call transcripts into blog posts?
Start with the pages you already have. Write down each repeated question, find the service, pricing or FAQ page that should have answered it, and fix that page in the caller's words. Draft a new post only when no existing page owns the question. Google's guidelines name mass-produced AI content without oversight as little-to-no-effort content, and a post for every call, published without real review, is the pattern that guidance describes.
Is it safe to put customer call recordings into an AI tool?
That depends on what your callers say, where the tool stores the audio, and what your industry requires. Medical, legal and finance calls carry details that should never reach a public page. Check your privacy obligations with whoever handles compliance, read the vendor's storage and training terms, and tell callers when you record. For content work, all you need is the written question with names removed.
How do I know which page a customer question belongs on?
Match the question to what the caller was trying to decide. Questions about what you do or whether a service suits them go on that service page. Questions about cost go on your pricing page. Practical questions about parking, referrals or cancellations go in the FAQ, and a new post is only for questions no existing page covers.
Your phone already tells you which pages are failing. A repeated question often comes from a customer who looked, found no answer and had to call. Fix the page they should have found, in their words, before you publish anything new. If your FAQ page feels finished, how many calls this week asked something it already claims to answer?