Monthly readings of AI answers for an engineering firm's long sales cycle

Monthly readings for a cycle measured in quarters

Between a buyer's first question and a purchase order there can be drawings, a quote, samples or a first article, a visit and a supplier audit, spread over months. An answer that names your firm sits at the start of that line, one step of a long customer journey. If an RFQ follows, it may arrive a quarter later, and by then nobody remembers the question that began it.

So the reading is not asked to find the RFQ. It records, month by month, what an answer said about your firm at a time a buyer might have asked, with the same questions every time so that each month compares with the last. A weekly reading leaves little time for anything to change except the answers' own run-to-run variation; a quarterly one lets several site changes pile up between two readings. Monthly is the cadence our methodology uses for monitoring, and the one this routine is built on.

Read the trend by quarter, not by month. A rise in one month can be variation; three months that agree, set beside a log of what you changed, are worth acting on.

Capability questions and questions with your name, scored apart

The routine runs two lists. The capability list is the bracketed panel from named for the part you can make: questions a stranger asks about a process, a material, a size or an industry, with no firm named. The name list holds the questions someone asks after hearing of you, from a trade show, a colleague or a directory.

Keep the scores apart. An answer to a question that contains your name can hardly avoid mentioning you, so the name list scores high whatever you do; mixed into one number, it lifts the total while nothing has changed for strangers. Read the name list the way brand monitoring reads any channel, for accuracy: every figure, certificate, address and contact the answers give, checked against your own pages. The mention rate is read from the capability list alone.

Fix name-list errors first. They reach buyers who already have your name, from a show or a referral, and a wrong certificate or an old RFQ address can end a conversation before it starts.

A change log, one dated line per change

A reading that moves means little unless you know what changed before it. Keep a log with one line for each change you make, on your site and off it, with the date it went live and the questions it should bear on.

Log what happened to you as well as what you did: a directory that rewrote your entry, a customer's press release that named you, a paper accepted for a conference. When a reading moves, the log shows what changed in the weeks before it; when a reading stays flat, the log shows which changes have not shown up yet. What goes on the capability pages themselves is in capability pages a buyer can compare.

Your inquiry log, read beside the record

The last source in the routine is yours alone: the inquiries you receive. Keep them wherever they live now, a spreadsheet or a CRM, with a few fields added:

• the date of first contact, and the date an RFQ followed, if one did;
• the capability asked about, in the buyer's words;
• any figure in the first message that you publish, such as an envelope or a tolerance quoted back to you;
• how the buyer says they found you, from a line on the inquiry form, "How did you first hear of us?", with an AI assistant among the options and a box for which one;
• the outcome: quoted, declined, lost or won.

Read it once a quarter beside the three monthly records and the change log. Count inquiries by capability, and set each count beside that capability's readings for the months before. An inquiry counts as coming from an answer only when the buyer says so; the rest is correlation, and belongs in the report labeled as correlation.

For Google's AI features there is one more source. Search Console's generative AI performance report shows impressions of links to your pages in AI Overviews and AI Mode by page, country, date and device, with no query dimension, so it can show whether a capability page appeared there, though not for which question. Add it to the quarterly read as a column per page; visits referred from assistants are covered in reading AI referrals in your analytics.

What no reading sees

A supplier choice in this trade runs through people and paperwork: an engineer who knows your work, a colleague's recommendation, a visit to the plant, a first-article inspection, a supplier audit, a place on an approved vendor list, a price. None of that appears in a reading of AI answers, and any part of it can outweigh everything the reading shows.

What the reading can claim is narrower and still worth having: whether a stranger meets your name for a capability you hold, and whether what someone with your name in hand reads about you is right, at the start of a process that ends elsewhere. It cannot tell you why an RFQ was won or lost, and nobody can promise that a better reading brings more RFQs. Report it that way to whoever asked for it; the one-page report for management shows a format.

The routine, month by month and quarter by quarter

Each month, on the same day

1. Run both lists on each assistant you measure, more than once per question; our methodology uses one assistant, Claude, and asks each question once per reading.
2. Record every answer with the columns set out in named for the part you can make, and paste the full text.
3. Add the month's changes to the log, on the site and off it.
4. Compare with last month, list by list: per capability, whether you were named, recommended and first; on the name list, every wrong fact.
5. Fix the name-list errors, then choose one capability page to improve before the next reading.

Each quarter

1. Read the inquiry log beside the three monthly records, the change log and, for Google, the generative AI report.
2. Write down which capabilities moved, which changes came before them, and which inquiries quoted a figure you publish.
3. Retire or add a question only with a date on its line, never by rewording one that already has history; the baseline measurement depends on it.

By hand, the routine is a spreadsheet and the hours it takes one person to read every answer. Measured for you, it is monitoring at $290 a month, which you can cancel in any month: twelve questions across both lists, written for your capabilities and put once a month to one assistant, Claude, with every answer kept verbatim and the visibility score set against last month's. Either way, the fixes happen on your pages, in the order the overview for machine builders and machine shops gives.