Track AI answers

Once a week, Mark asks each active buyer question in each market on six AI engines: ChatGPT, Gemini, Google AI Overview, Google AI Mode, Claude, and Perplexity. It stores each answer and checks two things: whether the answer names your business, and whether it cites a page on one of your websites.

Claude and Perplexity are asked through their APIs with web search on, not through the Claude or Perplexity app, so their answers can differ from what people see in the apps. The info icon next to each of their results says so. Weeks before Claude and Perplexity were added show them as Not measured.

Performance shows the results in its AI answers tab.

Before you start

• Add buyer questions in Research > Buyer questions, or right on the AI answers tab while it has none (see Suggested questions).
• To vet a question first, type it in the Question field on Research > Buyer questions and select Check AI volume. It shows the question's AI search volume for your priority market, like "1,200 a month in United States", before it is added. Nothing is saved.
• Each market you want checked needs a location. Set it in Strategy > Markets. A market without one isn't checked.
• AI answer checks are on by default. An admin can turn them off, or back on, in Settings > Workspace > Intelligence. The window that asks before turning them on shows how many checks each weekly run sends, like "12 questions × 6 engines = 72 checks". With more than one market that has a location, it counts markets too: "12 questions × 2 markets × 6 engines = 144 checks". A market without a location isn't counted, since it isn't checked.

Suggested questions

While your workspace has no active buyer questions, the AI answers tab lets you add them in place:

• Type a question and select Add.
• Or tick any of up to 10 suggested questions and select Add N questions. A question already on your list is skipped, and the tab says how many were added.
• Refresh suggestions writes a new list.

Each suggestion has a tag that says where it came from:

Up to 3 suggestions come from People Also Ask and Search Console. Mark writes the rest each time you open the tab. No suggestion repeats a question on your list, including ones you turned off. Nothing is saved until you add it.

If your Brand page is empty, you track no competitors, and there are no People Also Ask or Search Console questions, there's nothing to suggest from. The tab links to your Brand page instead. If Mark can't write suggestions right now, the tab says Couldn't suggest questions right now. with Try again. Typing a question still works.

After you add questions, the tab says what the first check covers, like "First check: 3 questions × 6 engines = 18 checks", the same count the intelligence window shows. Only markets with a location count, and with more than one the line names them too. With none, the tab tells you to set one instead.

Read the tab

1. Open Performance and select AI answers.
2. Pick a market in the filter bar under the tabs. With All markets, or with only one market, the tab shows your first market that has a location. Under the heading, Market names the market with its flag and Week of names the week shown. With All markets, the info icon next to Market says why only one market shows.

The date range and Compare don't change this tab. It always shows the latest weekly check that has finished. While this week's check is still running, the tab keeps showing last week's results and says so.

Other names for your business

An answer names you when it uses your business name, in any case, as a whole word. If AI answers call you something shorter, like "Verdella" for "Verdella Landscaping & Design Co.", add up to 5 other names in Settings > Workspace, under Business name > Also named. Select the pencil (Edit other names), type the names separated by commas, and select Save. Any member can change them.

An answer that uses any of these names counts toward your named rate and share of voice. The change applies to past weeks too, so the trend doesn't jump the week you add a name. Clear the list to go back to your business name only.

Engine tiles

Each engine gets a tile:

• Named: answers that name your business, out of the answers where Mark could tell. The tile shows the percent with the count under it, like "30%" over "3 / 10". See Other names for your business for what counts as naming you.
• Cited: answers that cite a page on one of your websites, counted the same way.
• When checks failed that week, how many succeeded, like "9 of 10 checks succeeded."
• How many questions got no answer that week, if any.

Point at the info icon by the engine name to see when that engine's newest answer was stored. On Claude and Perplexity it also says the answers came from the API, not the app.

Failed checks and questions with no answer don't count toward either total. When an engine has no answers to count, the tile says "No answers" instead of showing 0%, or "Not measured" when the engine wasn't asked that week.

Question table

Each buyer question gets a row, with one badge per engine:

Only Cited, Mentioned, and Not mentioned count. A failed check, no answer, an answer Mark couldn't check, or a missing check is not the same as "not mentioned", so those badges are grey and never count against you.

On a phone, each question becomes its own block with the engines listed under it.

Read one question's answers

Select a question row, or any of its badges, to open what each engine said. A badge opens the panel at that engine. The link in your address bar opens the same panel, so you can reload it or send it to someone on your team.

• Week: pick any earlier weekly check that asked this question in this market. The table behind the panel stays on the latest week.
• Each engine shows when Mark stored the answer, then badges:
  • You: cited, You: mentioned, or You: not mentioned, by the same rule as the table.
  • One badge per tracked competitor the answer named or cited, like Casita Builders: named, cited. If it named none, the panel says No tracked competitors named. Competitors you stopped tracking get no badge.
• The full answer, as the engine gave it. Your business name and its other names are highlighted in lime and each tracked competitor's name in blue, with a You and Competitor key above the answer. Highlights use your current business name and active competitors, so after a rename they can differ from a badge set when the answer was checked. Names in code, in a link's address, or split by formatting (half of the name in bold) are not highlighted, so a badge can say named with no highlight.
• Cited pages: every page the answer cited, by the title the engine gave it, with the site and path beneath. A page with no title shows the site and path alone, like greenco.com/pricing. Pages on your websites say Yours; pages on a tracked competitor's site show its name.

When an engine has no answer, the panel says why instead of showing badges:

Compare two checks

Changed a page and want to know if the engines noticed? Select Compare runs in the panel to put two checks of this question side by side. The button shows once the question has at least two finished checks in this market.

• Before and After: pick any two finished checks from the last 52 weeks that had one, by the time each one finished. A rerun in the same week is its own entry. Failed checks aren't listed.
• By default, After is the newest check and Before is the last check that finished before the question's answer page last changed: when it was published, or when Mark last pushed it to your CMS. The info icon next to Before names the page and the date. With no live answer page, Before is the check just before After. If the page changed after the newest check, the icon says no check has run since.
• Citations changed, per engine: Added lists pages cited in After but not in Before, and Dropped the reverse, each marked Your answer page, Yours, or a competitor's name. A count of pages cited in both follows. https://www.example.com/a/ and https://example.com/a count as the same page. When nothing moved, it says No citation changes.
• If one side has no finished answer from that engine (it failed, wasn't measured, or the market had no location), the box says which side, and citations for that engine can't be compared.
• Below the box, each check's answer sits in its own column with its time, badges, and full text. On a phone the columns stack.

The link in your address bar keeps both picks. Back to one week returns to the week view.

Baseline and wrong answers

Baseline. Under a badge, a grey line like was Not mentioned · Aug 31 shows what that engine said the first time Mark got an answer for that question in this market. It shows only when both the first answer and this week's badge are Cited, Mentioned, or Not mentioned, and they differ. Change the market and the line follows that market's checks. Failed, No answer, Not set up, Not measured, and Can't tell never get a line.

Wrong answers. When an AI answer gets you wrong (invented pricing, a product you don't sell, a competitor named instead of you), log it in the Wrong answers card below the question table:

• Log a wrong answer works for any engine, including ones Mark doesn't check weekly: Claude, Perplexity, Copilot, or Other. Engine, reason, prompt, what it said, and the date you saw it are required. Question, market, and a note are optional.
• The reason is one of Wrong fact, Competitor named instead, Negative framing, Outdated info, Missing citation, or Other. It's the first badge on each row. Wrong answers logged before reasons existed read Other.
• The weekly check files Competitor named instead by itself when an engine's answer names one of your active competitors and not you. Those rows show Found by weekly check, with the engine's whole answer as what it said. There's at most one open per question, market, and engine; once you resolve it, a later check that still names a competitor and not you files a new one.
• In a question's panel, Report wrong answer under an engine's answer opens the same form with the engine, prompt, date, market, and question filled in. Paste the wrong part of the answer.
• Open and Resolved list them newest first. Mark resolved moves one to Resolved with who and when; Reopen moves it back.
• A question row with open wrong answers shows a button like 2 open wrong answers. Select it to show only that question's; Show all clears it.

The card and the button count wrong answers for the market shown, plus ones logged with no market, including the ones the weekly check found. Anyone in the workspace can log, resolve, or reopen.

Share of voice

Below the question table, Share of voice in AI answers compares you with your tracked competitors in the latest weekly check. Pick All engines or one engine; the list and the chart both follow it.

• Share of mentions: the answers that name a brand, divided by the answers that name you plus each tracked competitor, added up. If ChatGPT names you in 1 answer and a competitor in 3, you see "25% of 4 mentions" and they see "75% of 4 mentions". It's a share among you and your tracked competitors, not a share of all AI answers.
• Named in: answers that name the brand, out of the answers with text.
• Cited in: answers that cite the brand's site, out of all answers.

Each answer counts once per brand, however often it repeats the name. An answer with no text counts toward Cited in only. Failed checks and engines not set up don't count at all.

Only competitors you track count. Track them in Strategy > Competitors; select a competitor's name to open its profile there. Past weeks use today's list, so a competitor you track now is counted in older weeks too, and one you stop tracking leaves every week. An answer names a competitor when it uses the competitor's name or one of its up to 5 Also named names, which you edit with the pencil on its profile. A change to those names applies to past weeks too.

The chart shows your share over the last 12 weekly checks. When a competitor leads this week, its share is the dashed line. A gap means no brand was named that week, which is different from 0%.

Cited pages

The Pages AI answers cite card lists every page the answers in the latest finished check linked as a source. Switch to Domains to group them by site. www.example.com and example.com are one domain; blog.example.com is its own.

Each row has a source:

• Yours: a page on one of your websites or its subdomains. These are the same answers the Cited badge counts.
• Competitor: a page on the site of a competitor you track. The badge names the competitor.
• Third party: any other site. These are the pages AI answers trust for your buyer questions. Each one is a chance to get your business onto that page or to publish a better page of your own.

Answers counts the answers that cite the row. On a domain row, an answer that cites two pages on that site counts once. Questions counts the buyer questions, and Engines shows the answers per engine. Filter by source with All, Yours, Competitors, or Third party. The first 25 rows show; select Show all for the rest. Links open in a new tab.

Suggested competitors

When the latest check cites third-party sites, the Suggested competitors card lists up to five of them, the ones cited in the most answers. Select Track to add a site to the competitors you track. It leaves the list and shows up in your share of voice after the page refreshes.

Not every cited site is a competitor, so track only the ones that sell what you sell. The card leaves out:

• your own sites, and competitors you track or stopped tracking
• any .gov or .edu site
• search and answer engines, such as Google and ChatGPT
• marketplaces and app stores, such as Amazon and the App Store
• review and directory sites, such as Yelp and G2
• reference, social, video, forum and code-hosting sites, such as Wikipedia, Reddit, YouTube, and GitHub

To track a site the card leaves out, add it on Strategy > Competitors. Tracking counts against the limit of 50 competitors. At 50, Track is disabled until you untrack one on Strategy > Competitors.

Answer pages

An answer page is a website post you wrote to answer a buyer question. Link one by picking the question in the post's Buyer questions field (see Create and submit a post). A page can answer several questions, and a question can have several pages.

Research > Buyer questions shows each question's answer pages in the Answer page column, with each post's status, or None when no page answers it yet. Select No answer page to see only the active questions with none. Sort by AI Search Volume to see which of those gaps buyers ask about most. Archived posts don't count as answer pages.

When you open a question's answers, the Answer page card lists its pages. For a live page, it says which engines cited that page in the week you picked and which didn't. Pick another week to see an earlier one. An engine whose check failed or didn't run that week is left out. In each engine's cited pages, your answer page shows Your answer page instead of Yours. A page that isn't published yet shows Not live yet: no engine can cite it until it goes live.

Group questions by topic

Give a buyer question a topic to see how you do on a whole theme, like "ADUs" or "modular homes". Set it when you add a question, or later from the Topic column in Research > Buyer questions: pick a topic you already use, type a new one (up to 60 characters), or clear it. Changing a topic doesn't change the question or any past check.

Once any question has a topic, this table groups under a header per topic, A to Z, with No topic last. Each header shows, per engine, how many of that topic's answers named or cited you, counted the same way as the tiles above. With no topics set, the table stays flat.

When checks run

Mark runs one check a week for your workspace. The first check starts on its own within an hour once AI answer checks are on, you have a buyer question, and a market has a location. After that, a check starts each week from Monday (UTC).

While a check runs, the tab says so. The question table keeps showing last week's results until this week's check finishes.

Questions you add after this week's check started are checked next week. The tab says how many are waiting.

To start this week's check sooner, an agent can start it over MCP with run_answer_engine_check. It follows the same rules: one check a week, so a second call returns the check that exists with its progress. If AI answer checks are off, it says so and starts nothing; an admin turns them on in Settings > Workspace.

Limits

• Six engines and one location per market. Claude and Perplexity answers come from their APIs, not their apps.
• Answers vary by user, place, and day. Each badge is one sample a week, not what every user sees.
• Mark matches your business name as a whole word and your site by its domain. An answer that names you a different way counts as not mentioned.

Ask an agent

The get_ai_visibility tool returns the same per-engine numbers for your first market with a location, or says why there's nothing yet, plus the share of voice list per engine, your share over the last 12 weeks, the top 10 cited domains, and up to 5 suggested competitors. The run_answer_engine_check tool starts this week's check, or reports its progress if it already started. The get_buyer_question_answers tool returns one question's answers per engine, with the same badges and cited pages, for any week, plus each engine's baseline and, for a failed check, the reason when Mark knows it. compare_buyer_question_answers returns the same two checks as Compare two checks, with the same default Before and the pages added and dropped per engine. list_wrong_answers and save_wrong_answer read, log, resolve, and reopen wrong answers. suggest_buyer_questions returns the same suggested questions as the tab, with their tags, and save_strategy_item adds the ones the agent picks. A new buyer question's reply includes firstCheck, the same questions × markets × engines count the tab shows. set_business_aliases sets the other names for your business, with the same limit and the same effect on past weeks as Also named. See the tool reference.

If the tab is empty