How to Track Campaign Visibility in AI Answers

To track campaign visibility in AI answers, start by writing down the three to six exact messages the campaign has to land, then add 10 to 15 tagged prompts that ask the questions the campaign wants to own, running nightly across ChatGPT, Gemini, Claude, Perplexity, and Google AI Overviews from at least two weeks before launch. Build the target outlet list around publishers AI can read, annotate your flight dates and placements so movement has a cause, and watch message pull-through weekly across the flight rather than checking once at the end. When coverage lands, group it and look at which placements get cited in AI answers; expect three to six weeks between a story running and any movement in the numbers. The result is a campaign report that says whether the story traveled into the channel where buyers are now asking their questions, not just where it was published.

Here's the moment this workflow is built for. The campaign wrapped, the coverage report looks strong, and somebody in the room asks what ChatGPT says about the company now. Nobody knows. The clip count doesn't answer it and neither does the impressions estimate.

What a coverage report leaves out

Traditional campaign measurement counts placements, reach, and sometimes sentiment. All of it describes what publishers did. None of it describes what AI assistants did with the result.

Those are separate outcomes now. A story can run in a well-known outlet and never enter an AI answer, because a lot of major publications block AI crawlers or sit outside the sources assistants lean on for your topic. Another story can run somewhere smaller, get read constantly, and quietly become the thing AI repeats about your category for the next year.

So a campaign has two scoreboards. The familiar one measures whether the media accepted your story. The other measures whether the machines that answer buyer questions accepted it. Tracking the second one takes setup before the campaign launches, which is the part teams usually skip.

The good news is that the setup uses things you already have. You have messages. You have a target list. You have flight dates. Each of those becomes something measurable.

This guide works with any AI visibility setup; the callouts show how each step runs in Spyglasses.

Step	What you're doing	What tells you it's working
1. Record the messages	Writing the claims down as text to match	Each message has a starting rate
2. Tag the prompts	Adding campaign questions to daily tracking	A filter that isolates campaign work
3. Score the outlet list	Checking which targets AI can read	Targets ranked by AI value, not just reach
4. Annotate the flight	Marking launch and placement dates	Every jump has a label next to it
5. Watch pull-through	Measuring message rates during the flight	Campaign messages rising, not just mentions
6. Group the coverage	Checking which placements get cited	A cited list and an ignored list

Write down the messages the campaign has to land

Every campaign has a small set of claims it's trying to install. The sustainability push wants people to know the packaging is recycled. The funding announcement wants the category framing to stick, not just the dollar figure.

Write those claims out as sentences, before launch. Three to six is the useful range. More than that and you're describing a brand, not a campaign.

Be specific about wording. "We're good for the environment" can't be measured. "Gear made from 80% recycled materials" can, because you can check whether an answer says something equivalent. Vague messages produce vague reporting, and they're the reason so many campaign retrospectives end with an argument about interpretation.

Add the negative ones too. If the campaign exists partly to correct something, record the claim you're pushing against, so you can watch it fade instead of hoping it does. Rebrands work the same way, and tracking a rebrand in AI answers covers that pattern in more depth.

Then get a starting rate for each message before anything launches. A message that's already at 30% is a different job than one sitting at zero, and after the campaign nobody will remember which was which.

Turn the campaign into prompts you can track

Your always-on prompts ask about your brand and category. A campaign usually pushes on something narrower, and if you only watch the standing prompt set, a real win can hide inside an average.

So add prompts that ask the questions the campaign wants to own. For a sustainability campaign, that's the questions a buyer asks about sustainable options in your category, not questions about you. Most campaign messages are meant to be picked up before anyone names a brand, so the prompts should be written that way.

A workable set looks like this:

• Four to eight problem questions, with no brand named, where you want your message to surface.
• Two or three comparison questions where the campaign's claim should differentiate you.
• One or two direct brand questions, to catch changes in how AI summarizes you overall.

Tag every one of them with the campaign name. That tag is what makes the reporting possible later, because it lets you filter the whole funnel down to campaign work and leave the rest of your tracking out of the picture.

Get them running before launch, not after. You want at least two weeks of pre-campaign data, or your before number is a guess.

Set a target outlet list you can score

The media target list is usually built from reach, relationships, and relevance. Add one more filter before you commit the outreach hours.

Can AI read the outlet, and does AI already cite it for your topic?

Those two questions split a target list fast. Some well-known publications block AI crawlers outright, so a placement there does nothing for this channel no matter how good the story is. Other outlets, often trade press and specialist sites, get read constantly and show up in AI answers about your category all the time.

That doesn't mean dropping the blocked outlets. A placement in a major paper still does the job it always did for credibility and reach. It means knowing, before the campaign, which part of your list can move AI answers and which part can't, so nobody expects both from the same placement. Building a pitch list AI can actually see walks through the scoring itself.

Once the list is set, attach it to the campaign as a goal with the outlets named. Then the question "did we get placements in the outlets that matter to AI" has a number instead of a debate.

Mark the flight dates and the placements

This step takes five minutes and saves the retrospective.

Mark the campaign launch date. Mark each wave if the campaign runs in waves. Mark every major placement on the day it publishes. Mark anything else that could move the numbers, like a product launch or an executive interview that isn't part of this campaign.

Six weeks later, when a line jumps, you'll be able to say what caused it. Without the marks you get a chart with movement in it and a room full of theories.

Annotations also catch the thing nobody plans for. A competitor announcement in week three, an unrelated news cycle, a change to your own site. Any of those can move your numbers, and if they're on the timeline you won't spend the retrospective crediting your campaign for someone else's event.

One habit worth keeping. Annotate the placement date, not the date you found the clip. Coverage tracking often lags by a few days, and a three-day error is enough to line a jump up with the wrong cause.

Watch message pull-through while the campaign runs

Now the campaign is live and the numbers start doing something. Two measurements matter, and they answer different questions.

Message pull-through asks how often your campaign claims show up in AI answers to problem questions, before any brand gets named. It's the closest thing to measuring whether the story traveled.

Share of voice asks how often your brand gets mentioned at all for the prompts you track. Filter it to campaign-tagged prompts and you get the campaign's own share, without your standing brand tracking diluting it.

Number	Question it answers	Where it moves first
Message pull-through	Is AI repeating the campaign's claim	Problem-stage prompts, no brand named
Campaign share of voice	Is the brand showing up for campaign questions	Comparison and decision prompts
Negative message rate	Is the claim we're correcting fading	Anywhere the old story is repeated
Citations from campaign coverage	Is our earned media being read	Sources listed under AI answers

Here's what a working campaign looks like in those numbers. An outdoor gear brand runs a six-week sustainability push behind the message "gear made from 80% recycled materials." That message starts at a 14% mention rate across 12 campaign-tagged prompts. Two weeks after the trade coverage lands, it's at 21%. By the close of the flight it's at 31%, and 5 of the 14 placements show up as cited sources in AI answers. The clip count for that campaign was 14 either way; the second set of numbers is the part that says the story moved.

Check weekly, not daily. Daily readings on a campaign look like static, because AI answers vary between sessions and one flat day means nothing. A weekly read across a six-week flight gives you a shape you can trust.

Expect a lag of three to six weeks between a placement running and pull-through moving. The story has to be published, crawled, indexed, and then chosen when someone asks a question. Teams that check a week after launch and see nothing usually conclude the campaign failed, when the pages hadn't been read yet.

If a message stays flat while others rise, look at the wording before you blame the outreach. Messages that are close to how buyers actually phrase the problem get picked up. Messages written in internal language tend not to, even with strong coverage behind them.

Group the coverage and see what AI actually cites

As placements land, put them in one group so the campaign's coverage can be measured as a set rather than one link at a time.

Then look at the part traditional reporting can't show you. Which of these placements is AI citing when it answers your campaign's questions?

You'll usually get three buckets:

• Cited and working. The placement shows up as a source in AI answers. This is the win, and it's worth noting the outlet because it's a priority target next time.
• Readable but ignored. AI can access the page and doesn't use it. Often the story is written for reach rather than for a question anyone asks, so nothing about it matches a query.
• Invisible. The outlet blocks AI crawlers or the page can't be read. It did other work for you; it just didn't do this.

That breakdown is the most useful thing to bring to the next campaign kickoff. After two or three campaigns you'll have a list of outlets that reliably convert into AI citations for your category, which is a better targeting input than reach numbers alone. Connecting earned media to AI visibility goes deeper on the valuation side of this.

Check the wider citation mix too, not just your own coverage. If AI keeps citing the same three sites when it answers your campaign's questions and none of them are yours, that's your pitch list for the next round, sitting in plain view.

What to keep after the campaign ends

Close the project and export the retrospective while the campaign is fresh. The export is more convincing than a summary written from memory, because the annotations are on the chart and the citation counts come with the prompts attached.

Three things are worth carrying into the next campaign:

1. The lag you observed. How many weeks passed between placements landing and pull-through moving. That's your planning number, and it beats any general estimate.
2. The outlets that converted. Which targets produced AI citations, and which produced coverage that AI never read.
3. The messages that traveled. The wording that got picked up tells you how buyers phrase the problem, which is useful well beyond PR.

Leave the campaign prompts running for a few weeks after the flight ends. Pull-through often keeps climbing after the outreach stops, as coverage gets crawled and indexed, and a campaign that looked flat on the close date sometimes looks different a month later.

Set the baseline before the next one launches. Almost every campaign that can't prove its AI impact failed at that step, not at the reporting step. Nobody wrote the messages down or ran the prompts before launch, so there was never a before to compare against.