The first time I saw my client’s brand in an AI answer, nobody had done anything to put it there.
I was working through one of the first exercises in an AirOps automation course, running a few prompts through ChatGPT, Perplexity, and Google AI Overviews to see what came back.
When I did, Perplexity mentioned CAT Electric Vision, a small firm that sells earthing and lightning surge protection equipment on the Romanian market, right in its answer. In ChatGPT, the same brand only showed up as a buried link in the sources panel.
The firm has been selling for three years, had a team with over a decade of experience, and I know it well because the client is a close relative. Like a lot of small, local businesses, they’d been doing fine on referrals, returning enterprise clients, and being found online. Content marketing came in bursts: an initial push to set up brand assets and a content strategy, then business would pick up, the online presence would slow down, and getting back to it meant a reconnecting phase each time.
Even so, the brand was already appearing in AI search without anyone working on it. It got me thinking: if it showed up organically, then with more effort behind it, maybe it could show up more often and across more engines?
That’s what AEO (answer engine optimization) comes down to: working on how often AI engines mention you when someone asks a question you could answer. It’s something most marketers don’t have a system for yet, and neither did I.
I’m a generalist content marketer, and this pipeline was my first automation build. But I’m curious; I look for chances to upskill, and thanks to my course, I knew how to use AirOps, a content optimization tool. So I followed the thread: if people are asking AI engines questions the brand could answer, how do I find those questions?
What I ended up with is a content loop that runs about once a week. It checks what the brand has already published and how it performed, finds the questions where the brand is invisible in AI answers, checks whether the company has enough material to credibly answer them, and turns the best ones into writer-ready LinkedIn briefs that land in Buffer.
Here’s how I built it (and what I’d keep in mind if you want to try something similar).
Why LinkedIn was the obvious place to start
Around the time I was running those prompt exercises, I kept coming across research on where AI engines pull their answers from, and LinkedIn came up again and again.
Semrush found that LinkedIn is the second-most-cited source across ChatGPT Search, Google AI Mode, and Perplexity, appearing in 11% of AI responses on average. For professional queries specifically, Profound found LinkedIn is the most cited domain across all six major AI platforms, including ChatGPT, Gemini, and Copilot.
That second stat mattered most for this client. The people asking about surge protection are engineers, installers, and building owners — professional queries by definition. There’s even data on what gets a LinkedIn post cited: Scrunch found that technical details raise the odds by 77% and named entities by 33%, while Unicode bold formatting lowers them by 58% on ChatGPT specifically. A small firm with deep technical expertise is well positioned for exactly that kind of post.
The firm also had a head start. They were already using LinkedIn to reach clients and had older material to build on. And LinkedIn posts take less effort to write than blog articles, which is ideal for a small team whose content marketing happens in bursts.
One decision I made early was to track brand mentions rather than citations. A citation is when an AI engine links to your post as a source; a mention is when it names your brand in the answer itself. Peec AI, the visibility tool I’ll get to shortly, tracks the two separately. For a firm whose clients ask questions like “recommend a lightning protection system supplier in Romania,” being named in the answer is a huge win.
📚 Why LinkedIn Is the Most-Cited Source in AI Search (and What Your Business Should Do Next)
The stack I landed on
The finished loop runs on four tools, each with one job:
AirOps orchestrates the whole workflow and holds the company’s Brand Kit and Knowledge BasesPeec AI tracks which AI prompts mention the brand, and which don’tBuffer supplies the LinkedIn engagement data and is the editorial home where finished briefs are sent for the writer to review
Here’s why I chose each tool.
AirOps, the orchestration layer
AirOps was my introduction to automation, through the same education cohort where I ran those first prompt exercises. It can be an expensive tool, but the trial period covers an experimentation phase like this one, and it comes with an AI agent called Quill. I created the workflow by conversationally sharing my ideas with Quill and having it sort out the technical parts, which made the whole thing much less intimidating for a first build.
AirOps also helps you create a Brand Kit and Knowledge Bases inside the platform. Once the company’s product pages, social posts, and YouTube transcripts were in one place, I became more aware of how much unused information the firm had, and I started exploring ways to turn it into content ideas.
Peec AI, for visibility data that covers the Romanian market
The deciding factor here was language. AirOps has its own Prompts feature, but it didn’t support Romanian yet when I built the workflow. Scrunch focuses on English prompts too, and Profound and AthenaHQ, at first glance, are built for enterprise teams rather than a small firm experimenting.
Peec AI showed up in my LinkedIn feed one day, and it turned out to work well for the Romanian language and market I had in mind. They’re headquartered in Germany, which might explain the local-market coverage, and there’s a trial period as well.
Buffer, for engagement data and as the editorial home
The workflow only works as a loop if it can see how the last round performed. Buffer is a LinkedIn marketing partner, so engagement data for posts published through it comes back through Buffer’s own API, the same one the workflow was already talking to.
Buffer picked up the second job once I saw that post ideas can land in the Create board through the API. I didn’t want the writer tied to the workflow or needing AirOps access, and Buffer solved that without adding another layer like a Google Doc or a Notion board.
The Create board is structured like a kanban (a board of columns where work moves through stages), which matches how I’m used to seeing editorial work organized. Multiple writers can have access, an editor or the owner can approve posts before scheduling, and all of it stays separate from the automation itself.
How the workflow runs, end to end
The workflow runs once a week. Each run opens by checking what already happened — what got scheduled, what got posted, and how it performed — before deciding what to make next. From there, it works through four steps.
Step 1: Check what’s already in motion
The AirOps Playbook agent connects to Buffer through the API to see what’s already drafted or queued, so it doesn’t create duplicate ideas. It also checks its own internal Grid, the workflow’s memory of previous runs. Then it pulls engagement data for published LinkedIn posts through the same Buffer API, reads the dataset that comes back, and extracts reactions, comments, impressions, and reach for each post. Posts that performed well become signals for follow-up content.
This step replaces a person having to open the apps themselves or message team members to get up to date.
Step 2: Ask where the brand is invisible
Next, the agent connects to Peec AI through a token and pulls data in two passes. The first brings back the prompts themselves: text, topic, volume, and tags. The second brings the measurements: visibility, share of voice, sentiment, position, and how competitors show up for the same prompts.
Peec runs the tracked prompts daily across three AI engines (ChatGPT, Perplexity, and Google AI Overviews) and records whether the brand is mentioned in each answer, so a prompt’s visibility score is the percentage of answers that mention it. The workflow flags a prompt as low visibility when it sits at 0% after at least seven days of tracking, drops under 20%, or falls 15 points or more in a month.
🖊️ In practice, the thresholds don’t filter much yet, since the brand’s overall visibility is 7% and nearly every prompt qualified at 0%. But this step will matter more as visibility grows.
In a nutshell, this step answers the question “Where are we invisible?” It replaces a marketer manually going through the prompts, doing data entry, and exporting CSVs.
Step 3: Map the gaps against what the company can credibly say
For each prompt, the agent searches the three Knowledge Bases: 390 older product pages, the company’s social media posts, and YouTube transcripts. It checks whether the company has enough material to credibly answer each question; prompts with no supporting evidence get dropped, since without material to draw on, the idea would be hollow.
Surviving ideas get ranked on four factors: the size of the visibility gap, how often people ask the question, the strength of the Knowledge Base evidence, and whether competitors already own the answer space. Only the top five move forward, a limit I set on how much we can work on at once.
Then the agent pulls the Brand Kit rules (the LinkedIn post format, persona, audience profiles, and writing rules) and combines them into a single self-contained brief per prompt: topic, angle, audience, key points, brand positioning, verbatim source material, and format guidance.
Step 4: Land in Buffer as writer-ready briefs
The agent sends each brief to Buffer with a single API call, and it shows up in Ideas, inside the Create space. When the writer opens Buffer, an idea looks like this:
Post idea: the topic, angle, which audience to write for, and a proposed day to schedule. For the idea in the screenshot, the agent targeted investors and building beneficiaries, flagged it as high priority, and proposed a Wednesday slot.AEO prompts targeted: the two Peec prompts the post is meant to answer, both at 0% visibility with medium volume.Key points: the substance of the post. This one suggests explaining that “installation included” is a marketing phrase, not a certification, and that a client should ask who designed the system and whether the project complies with the technical standard.Brand positioning: how to position the company as the answer: the certification stack, 2,500+ completed projects, and a manufacturer whose products are independently lab-tested.Products to name: two specific lightning rods.Source material: excerpts from older LinkedIn posts pulled from the knowledge base, so the writer draws on real product language.Post format: the Brand Kit rules are 50 to 299 words, no Unicode bold, link in the body, direct address, open with a direct statement. The no-Unicode rule comes straight from the earlier citation data.
On the next run, the agent checks whether the post went live, pulls its metrics from Buffer, and checks Peec again to see whether the brand is now mentioned for that prompt. That check is what closes the loop.
What’s changed so far
It’s too early for impressive numbers. The posts created from the pipeline’s briefs are just going live, so there’s no change in cadence or publish rate to report yet.
What has changed is that where my ideas used to stall in a brainstorming phase, they now arrive in Buffer as structured briefs, and the writer spends less time reviewing. We’re not posting four times a week yet, but we have the capacity to.
On visibility, four of the five prompts the pipeline selected are no longer at 0% in Peec, and the posts weren’t live yet, so I can’t attribute the change to the pipeline. Not every visibility shift can be traced to new content, and it’s useful to know that before you promise a client otherwise.
The clearest result so far is buy-in. Before this, AI search wasn’t on the owner’s radar; their focus was compliance, EU rules, and email campaigns. The experiment put AI search on the list, and that’s a result in itself.
What I’m watching for next is whether the published posts move their prompts off 0%, and whether ideas from the pipeline make it to publish at a higher rate than ideas did before.
If you want to try something similar
You don’t need this exact stack. The lighter-weight version looks like this:
Run the discovery exercise yourself. Ask ChatGPT, Perplexity, and Google AI Overviews the questions your clients ask you, and note where the brand shows up and where it doesn’t. The two questions to focus on: how is the brand mentioned in AI searches, and where are the engines failing to mention it when someone asks something relevant?Centralize what the company can credibly say. I used AirOps’ Brand Kit and Knowledge Bases, but a set of Notion pages or Google Drive docs works too: a company description, a product and service list, expertise areas, and your LinkedIn post format rules. With a Notion or Drive MCP connector, an LLM like Claude can pull from those docs and handle the orchestration.Get prompt data, however it fits your budget. An AI visibility tool saves the manual work if it covers your language and market. If it doesn’t, you can gather prompts yourself, for example with a Chrome add-on like ChatGPT Search.Score the gaps with four questions. Are we invisible for this prompt, do people ask this type of question, can we back up an answer with real material, and are competitors already there? Drop anything you can’t support with evidence.Put the output where your writer already works. For me, that’s Buffer’s Create space through the API, so the review and publishing step needs no access to the automation at all.
Ready to build?
If you’re taking Buffer’s API for a spin, we’ve got resources to get you moving. Our developer docs cover the GraphQL schema, auth flow, and quick-start examples. The Buffer MCP server docs walk through plugging it into Claude or any MCP-compatible AI agent.
If you need hands-on help, our support team is around, or you can join our Discord server and chat to other people building with the API.
We’d love to hear about what you make. Find us in Discord, or @buffer on all major social channels.
