Recommendations

Being named in an AI answer is not the same as being recommended. An answer can single you out as its pick, list you among several options, mention you in passing, or tell the buyer to look elsewhere. Share of Voice counts all four the same way, because each one names you. Recommendations tells them apart, so you can see how often AI picks you, who it says you are best for, and the reasons it gives when it steers buyers away.

What You'll Learn

• Which answers are classified, and how
• The four categories, and how caveats differ from them
• What the dashboard card shows
• The two families of numbers on the Recommendations page: rates and shares
• How to get from any number to the answers and the exact words behind it
• How the page changes for a category or person property and in the client portal

Which Answers Are Classified

Recommendations reads the answers from daily tracking, where a project runs your prompts every day. Within those answers it uses discovery prompts, the questions buyers ask when they are looking for options. Brand-identity and comparison prompts are left out, because a prompt that names you invites an answer about you. AI Visibility Reports are not part of it.

For every answer, an AI judge classifies every brand the answer names, whether you track it or not. That is what makes the market view on this page possible, and it is why a brand you start tracking later arrives with its history already in place.

If the property does not have daily tracking yet, the page shows Nothing classified yet. Set up a project, and classification starts with your next daily run.

The Categories

Each brand named in an answer gets exactly one category:

Top choice and one of many together count as recommended.

Two more values complete the picture when you look at one brand across answers:

• Not mentioned: the answer does not name the brand at all.
• Unclassified: the answer names the brand, but it was not rated. This is rare.

Caveats are separate from the category

A caveat is a reservation an answer attaches to a brand. Each caveat records three things:

• Reason type: Price, Complexity, Company size fit, Missing capability, Integrations, Support and service, Reliability and quality, Security and compliance, Reputation, or Other.
• Reason: what the answer said, such as "larger companies outgrow it".
• Audience: who the caveat applies to.

A caveat does not change the category. A top choice can still come with a caveat such as "larger companies outgrow it": the answer picked the brand, and it also said who it would not suit.

That split shapes how the page reads. The Why AI cautions against you section counts only the caveats on answers that cautioned against you. Caveats on answers that still recommend you stay with those answers, as chips on each row of the answers page.

The Dashboard Card

The How AI recommends you card on the Brand Dashboard covers the last 7 days, like every dashboard card. It shows:

• Your top-choice rate, with the change against the previous 7 days
• A bar showing how every answer that week treated you, from top choice to not mentioned
• One line on cautions: how many answers cautioned against you and the reason given most often, or "No cautions this week."

The Recommendations link on the card opens the full page.

The Recommendations Page

Filters

Every number on the page links to the answers behind it, so you can check any rate or share against the words AI used.

Two families of numbers

The page keeps two kinds of numbers apart, because they answer different questions.

Rates are per answer, for one brand. They ask how often AI treated you a certain way in the answers in this slice. The denominator is the number of answers in the slice, and changes compare with the previous period of the same length.

The rates sit in the How AI treats you section, headed with your brand's name. Below them, a bar shows every answer in the slice by how it treated you, and it adds up to the number of answers.

Shares are across the market. They ask how much of what AI recommended went to each brand. The denominators include every brand the answers named, including brands you don't track.

Because every brand is in the denominator, each share column adds up to 100%, including Other. Rates for different brands do not add up to anything, because one answer can recommend several brands.

Here is how the two relate. In 100 answers, AI picked you as its top choice in 20 and recommended you as one of many in 40. Your top-choice rate is 20% and your recommended rate is 60%. Those same 100 answers made 400 recommendations across every brand they named, so your 60 recommendations give you a share of recommendation of 15%. The rate tells you how often you make the list; the share tells you how much of the list is yours.

Why AI cautions against you

This section lists the reasons AI gave when it steered a buyer away from you. Reasons are grouped by reason type, with a count and a share of all cautions, and each one says who it applies to. Reasons that are new since the previous period are marked New. Each reason links to the answers that gave it.

Only cautioned-against answers count here. A caveat on an answer that still recommended you appears as a chip on that answer in the answers page instead.

Where AI picks you

The buyers and use cases AI names when it recommends you, taken from the best-for notes on your recommendations. Each phrase shows how many recommendations named it and how many of those were a top choice. A phrase is marked Rising when it appeared in at least 20 more answers than in the previous period, and a quarter more, and New when it did not appear in the previous period.

Cited, not recommended

This tile counts answers that cite one of your own pages but leave you out of the recommendations: not named, named in passing, or cautioned against. It also shows what share of the answers citing you that is. AI used your page as a source and did not recommend you.

It is computed from the citations Spyglasses already stores, and the tile links to exactly those answers.

Share of recommendation

This section ranks every brand AI recommended in the slice. Each row shows the brand's share of recommendation, its top-choice share, its cautioned rate (a per-answer rate, not a share) and how many answers named it. Your row is marked You, and brands you don't track are marked Not tracked. Brands beyond the top of the list fold into Other; Show every brand lists them all.

Under the table, How AI describes each brand shows, for each brand, how the answers that name it treat it, strongest first.

Tracking a brand AI recommends

A brand you don't track has a Track action in the list. Tracking it adds the brand as a competitor, and you enter its website. Its recommendation history is already here: every brand in every answer was classified from the start, so nothing needs to run again.

Over time

The trend chart plots your top-choice rate and cautioned rate by day or by week. Hover a point to see how many answers are behind it.

Early reads

When a slice holds fewer than 30 answers, the page marks it as an Early read. At that size one answer moves a rate by several points, so read the counts rather than the percentages until the slice fills out. Widening the window or clearing a filter gives you a steadier number.

The Answers Page

Recommendation answers lists one row per brand per answer, newest first. Following a number from the Recommendations page lands you on exactly the answers behind it, with the filters already set.

Each row shows:

• The category
• Any best-for notes
• Caveat chips, each with its reason and who it applies to
• The exact quote from the answer that the rating rests on
• A Cites your site marker when the answer cites one of your pages

Filtering the answers

The page keeps the same window, project, stage, platform and prompt tag filters as the Recommendations page. Under Rows, four more narrow the list:

The answer record

Opening a row takes you to the full answer record. Its How AI rated each brand in this answer panel lists every brand the answer names, with the category it earned, who it is best for and any caveats. The brand you came from is selected, and the words behind its rating are highlighted in the answer. Select another brand to move the highlight to its rating.

Export

Export CSV writes one file with every matching row plus each answer's full text, so the ratings and the words behind them stay together. A very large export holds the newest 10,000 rows; narrow the filters to export the rest.

Category Properties

A category property tracks a market rather than one brand. It gets the same page under the name Who AI recommends, with three differences:

• There is no "You" row.
• The market at a glance takes the place of your rates: the share of answers with a top pick, the share with a caution, how many brands AI named, and how many brands one answer names on average.
• Why AI cautions against brands here covers the cautions on every brand in the market.

Person Properties

On a person property the page works the same way with people in place of brands. Every name in every answer is classified, including people you don't track.

In the Client Portal

In the client portal, the page is read-only. Clients can read every number and open the answers behind it; the Track action is not shown.

Availability

Recommendation categories are included with daily tracking on every plan, at no extra charge.

Best Practices

Start with the cautions

A caution names a reason and an audience. When the same reason keeps coming up for the same audience, you have a specific claim to answer on your own pages, in earned coverage, or in the product itself.

Read rates and shares together

A rate tells you how often you make the list; a share tells you how much of the list is yours. If your recommended rate holds steady while your share of recommendation falls, AI is making more recommendations to other brands in the same answers.

Check where AI picks you

When AI recommends you as one of many, the best-for note is how it tells the options apart for a buyer. Low best-for coverage means AI lists you without saying who you suit. Compare the phrases in Where AI picks you with the audience your key messages are aimed at.

Work the cited, not recommended answers

These answers already use one of your pages as a source. Read the answer next to the page it cited and check whether that page makes a clear case for who you are best for.

Track the brands AI keeps recommending

If a brand you don't track keeps showing up near the top of the list, track it. Its history comes with it.

Related

• Brand Dashboard: Where the How AI recommends you card lives
• Projects: Turn on daily tracking, which Recommendations reads
• Competitors: Manage the brands you track, including ones added with Track
• Prompts: Author the discovery prompts that feed this page
• Key Messages: Define the positioning you want AI to repeat
• Tracking a Category: Measure a whole market, where this page becomes Who AI recommends
• Client Portal: Share the page with clients, read-only