An answer engine does not describe your business. It describes the residue your business has left on the web — and for a firm that has traded here for forty years, that residue is thick, scattered across other people's websites, and mostly out of date.

A layer has been added on top of search rather than replacing it. Underneath, the machinery runs as before: pages fetched, stored, scored, ordered. On top, a model reads that material and writes a short passage, and the person who asked frequently stops reading there.

What follows treats that layer at the level an owner here can act on: why being named is a different event from holding a position, which documents get used, how a visibility figure is built, and how to report it without writing a sentence you will have to defend a year from now.

Format · What the asker receives

A paragraph handed over, instead of ten links to sort through

A results page is a set of candidates: ten entries, ads, a map block, and a reader deciding which two to open. The entry sitting seventh is still in play, because the reader does the choosing and sometimes chooses the seventh.

A composed answer removes that step. Three or four sentences, two or three sources attached, and the choosing already done by something that will not explain its reasoning. Either the firm is inside the passage or the exchange finishes without it.

The difference stated plainly. A ranked page competes for a reader's attention. A cited source has already survived a filter the reader never sees. Both are fed by the same index, which is why this is a second outcome of one body of work rather than a second discipline.

Not every question changes hands this way. Somebody weighing roofing materials wants to browse. The owner of a Chestnut Hill twin with a failing slate roof wants one name and a number, and calls whichever arrives first — a minority of searches, and a majority of the ones that turn into work.

Mechanics · Two different verdicts

A position is a place in a line; a citation is a yes or a no

Rank is continuous. You sit at six, the thing at five is visible and specific, and closing that gap is defined work. Even eleventh place is informative: you are in contention, and roughly this far off.

Citation has no such gradient. There is no almost. A firm can hold the strongest position on the page and still be absent from the passage above it, because that passage was assembled from three documents stating something checkable while yours said the team is passionate about craftsmanship.

QuestionClassic positionNamed in an answer
What is the outcomeA number between one and one hundredInside the passage, or not
Where the record livesServing logs and rank trackersNowhere anybody publishes
How it changesGradually, in viewAbruptly, with no notice
What decides itRelevance, links, quality of the pageClarity, corroboration, how recent the sources are
Same result twiceBroadly yes, from a neutral checkNot reliably, even an hour apart

The last row does most of the damage to expectations. Two people in one office, phrasing a question two ways, get two different sets of companies.

Philadelphia · The archive problem

An old business has an old paper trail

Here the subject becomes specifically Philadelphian. A model describing your company draws on whatever the web has accumulated about it, weighted toward whatever several sources repeat. For a business founded in 1984 that has rebuilt its website three times, that accumulation is enormous and almost none of it is recent.

Your current site is one document. Against it stand a listing on a directory service that folded in 2011, a review profile still naming the founder's son although he sold in 2019, a supplier's dealer locator with the address from before the move to the Delaware County shop, and a news item from 2014 about a discontinued product line. The machine's version of the company is frequently the stalest version available, because staleness and repetition look identical from outside.

Residue

The directory that outlived its company

A listing on a local directory that stopped taking edits years ago and never went offline.

  • Old address, old hours, old phone
  • Still crawled, still quoted
Residue

The review profile with the previous owner

A profile naming whoever ran the business two owners ago, with reviews to match.

  • Ownership changes rarely propagate
  • Outside pages carry weight yours does not
Residue

The news item from 2014

Genuine coverage, accurate when published, about a company that no longer exists in that form.

  • Permanent, credible and wrong
  • Nothing to retract, and you would not want it retracted
Residue

The association page nobody updated

A chamber, a trade body or a licensing roster listing a contact who retired.

  • One email usually fixes it
  • Nobody has ever sent that email

Three regional conditions deepen the pile. Businesses are old, so more of them has been written down. Ownership passes within families and between neighbors, so one trading name covers several companies over time. And the domain has usually changed once: a firm that traded under the founder's name for thirty years, rebranded in 2008 and kept the old domain as a redirect has two names in circulation, each with its own trail.

The resulting description is therefore not a verdict on your marketing. It summarizes what others wrote about you before your last two rebuilds, and stays that summary until newer material outweighs it.

Retrieval · What gets quoted

Which documents end up inside the answer

Nobody outside the model builders can list the retrieval rules, and any confident account of them is sales material. What is observable is which kinds of pages keep appearing as attributed sources — consistently enough to plan around.

  • Sentences that finish. A paragraph naming the trade, the counties covered, the founding year and the licenses held can be lifted whole. A hero image reading Trusted cannot.
  • Particulars with numbers attached. Response times, license numbers, service radius, years in business. Verifiable specifics survive the trip into a composed paragraph; adjectives fall out on the way.
  • Independent sources that agree. One address and one specialty stated by a licensing board, a membership roster and a local publication counts for more than the same claim made three times on your own domain.
  • One spelling of the company. The same name, the same address format, the same phone number wherever the business appears. Three near-variants read as three thinly documented firms.
  • Material nobody has to trade an email for. Capability sheets locked in gated PDFs contribute nothing, and old firms keep much of their best content in exactly that condition.

None of this is new. It is the clarity that always helped a human buyer, minus the tolerance for vagueness: a buyer could infer that a company was serious, and retrieval cannot infer, only quote.

Cleanup · On land you do not own

Correcting a record that lives on other people's websites

Stated that way, the work is obvious and unappealing: most of what needs correcting is not on your server. It is a cleanup across directories, associations, publishers, suppliers and review platforms, each with its own process or none at all.

Begin with an inventory. Search the company name, the founder's name, the old brand, the previous address and the old phone number, and list every page returning something stale. Most firms stop counting around twenty.

Where the stale record sitsRealistic outcomeEffort
Association or chamber rosterUsually corrected within a week of askingOne email
Licensing or regulator listingCorrected when the underlying filing isA form, sometimes a fee
Review platform with an owner fieldClaimable, then editableVerification, then an hour
Supplier or manufacturer locatorDepends entirely on your repA phone call, then patience
Directory with no staff leftFrequently nothing at allNot worth chasing
Archived news coverageStays as published, correctlyOutweigh it, do not fight it
This cannot be fully controlled, and pretending otherwise wastes money. Much of the record about an established company sits on sites with no contact address, no editorial staff and no reason to answer you. Correct whatever still has an owner, publish enough current material that the current account becomes the best-corroborated one, and there the influence ends. Any offer to clean up how a model talks about your company promises an outcome the seller does not control.

The productive framing is displacement, not deletion. Old material loses weight when newer material is more specific and outside sources repeat it. That is why link building belongs beside this work: placements across a partner network exceeding 230,000 sites put a current account of the business into circulation on properties that are not yours.

Worth doing

The half you can influence

Records with an owner, a form or a human being behind them.

  • Claim and correct the review profiles
  • Update every roster and membership page
  • Restate current facts as plain text on your own site
Let it go

The half you cannot

Records that are abandoned, archived, or accurate about a company you no longer are.

  • Dead directories with no contact route
  • Coverage correctly reporting 2014
  • Forum threads and scraped copies of copies
Estimates · Behind the number

Where a visibility score comes from, and what it can carry

Because nothing is published about who gets named, a visibility figure has to be manufactured, and the method is worth knowing before reading the output of the AI Analytics section of the panel. Put a structured set of questions to a model, note which domains come back, repeat, reduce the tally to a score. That is sampling — a respectable technique, and not measurement.

AI Analytics · Market research

How a model frames the market you trade in

Six views for an owner deciding what to write before writing it.

included in the panel
  • Competitiveness score and market circle. Rival firms sorted into three rings — leaders, middle, niche. The second ring is usually where a family business learns something.
  • Written market context. Where the domain stands, roughly what traffic it draws, and what the model reads as unclaimed.
  • Queries grouped by intent. What people are trying to accomplish, which is where a browsing question separates from a hiring question.
  • Pages worth working on. URLs flagged for expansion or internal linking, with competitor strengths and gaps beside them.
  • A single landscape value. One figure for standing across AI search, readable as a trend line and nothing finer than that.
6
research views
3
rings in the circle
1
figure, many samples
No engine publishes a citation count. No feed anywhere reports which sites an answer engine named, how often, or for whom, and no vendor has privileged access to one because none exists. Every figure in this category — this panel's included — was assembled by putting questions to a model and tallying what came back. Sampling described in the vocabulary of measurement is still sampling.

That does not make it worthless. Asked the same way month after month, a fixed question set shows whether the direction is right and whether a quarter of writing shifted anything. Direction is the defensible claim; a percentage is not.

An inferred score is not a measured metric. Impressions and clicks are counted events that reconcile against Search Console. A visibility score is the output of a sampling routine. Formatted alike and stacked in one column, the two borrow each other's authority, and the borrowing always runs toward the weaker number. Keep them in separate blocks, and print the method under the estimate.
Publishing · Plain facts

What to write down, sector by sector

The work is dull and cheap: typing what the staff already know. A workable test is whether a stranger can lift three verifiable facts off a page in ten seconds.

SectorKnown, but never publishedWhere it belongs
Trades and constructionTownships licensed in, permit experience, lead timesLocation pages naming the permitting authority
Legal and professional servicesJurisdictions, matter types, typical engagement sizeA practice page written in nouns, not promises
Healthcare practicesSpecialties by site, languages spoken, plans accepted, affiliationsOne page per location, dated and maintained
Life sciences suppliersCertifications, tolerances, minimum orders, validation papersA capability sheet in HTML, not a download
Food and hospitalityKitchen hours, private dining capacity, catering radiusA page kept current, not a PDF menu
Neighborhood retail and servicesBlocks served, parking, delivery area, year openedThe about page, in its opening lines

Two habits do most of the harm on old sites. The gated PDF, which locks a supplier's most quotable document behind an email address nobody follows up. And the founding-story page in a 1990s register: three paragraphs of pride with no date, place name or number in them.

One hour, two payoffs. List the five questions the front desk fields by phone each week. Answer each in three plain sentences, with numbers where numbers exist, and post them as text on the page they belong to. The ranked page improves and the odds of being quoted improve together.
My SEO · Entry tier

AutoSEO — the gaps get found for you

For an owner who wants that writing list produced without putting somebody on the payroll to produce it.

$149 per month · per domain
  • Suggestions aimed at your own URLs. Recommendations against the pages you have, not a checklist any agency could hand across a table.
  • Terms found and ordered automatically. Candidates come from the connected property, live results pages and any seed words you add.
  • Placements running in the background. The partner network supplies outside corroboration at a volume no owner arranges by hand.
  • Counted data in the same window. Search Console figures and rank tracking sit next to the campaign, with a chat assistant reading the project's numbers.
$149
per domain, per month
230,000+
partner sites
4–8
weeks before anything moves

Two layers do not require two budgets. What makes a page usable to a customer makes it usable to a retrieval system, and a trade-press mention lifts the ordinary result too. Charging separately for the new layer invoices one job twice, which is why the research views sit beside campaign management.

Reporting · What survives a challenge

Putting the estimate in a monthly report honestly

The instinct is to lead with the number. Do not. It belongs in the report, in a fixed position, with one line saying where it came from — the difference between an indicator and something read back to you later.

Keep this out of a hospital system's marketing review and out of a board file. A vendor-sampled estimate has no place in a governance pack, a credit application, a valuation exhibit or any paper signed in a fiduciary capacity. Those documents need numbers with a traceable origin; this one has a method, not a provenance. Its home is the working report a marketing team uses to pick next month's writing.
  • Label it wherever it appears. Estimated by sampling, with the date and the sample count. An unlabeled score becomes a fact the moment somebody pastes it into a slide.
  • Give the trend, not the level. Rising across two quarters is supportable. A precise share of answers is not.
  • Never on the revenue page. Place it with the content work it responds to, at a distance from bookings and forecasts.
  • Look at it four times a year. Week-to-week noise exceeds most real change, so frequent readings buy meetings about variance.
quarterly
sensible reading interval
10,000
rows in CSV or JSON
250
rows in a PDF
28 / 90
day comparison presets

Mechanically it is easy: the configurable report builder sets the estimate beside counted metrics under your own logo. The discipline is not mechanical. Label the figure, or keep it out.

Common questions

Should we stop paying attention to ordinary rankings?

No — the dependency runs the other way. Composed answers are built out of the same index, so a page conventional search cannot find will not be quoted either. Rankings are also the only part of this subject supported by counted data.

The description we get back is years out of date. How do we correct it?

Start on your own site, then accept that most of the job is elsewhere. Inventory every outside page carrying the old address, owner or brand, and fix those with somebody behind them: rosters, claimable profiles, dealer locators, regulator records. The rest is displacement — publish current specifics and get outside sources repeating them.

We asked three times and got three different companies. Is something broken?

Nothing is broken; the layer behaves that way by design. Phrasing, timing and the rest of the conversation all shift what comes back. One check proves nothing either way, and any number attached to this has to rest on repeated sampling.

Our old brand name still circulates. Is that helping or hurting?

Some of both. The former name carries decades of outside corroboration and is often better documented than the present one, which is why it keeps surfacing. Hold the redirect, put the change in plain text on the live site — former name, year, same company — and give outside sources something current to repeat.

How soon does writing show up at this layer?

Slower than in rankings, and less regularly. A page must be crawled, kept, then chosen as a source, and only the first two steps are visible to you. Reckon on four to eight weeks before the counted metrics stir.

Three conclusions survive the noise. The account a model gives of an established Philadelphia business is assembled largely from pages other people own, most older than your website. Correcting that record is slow inventory work with a hard ceiling on how much you control. And every number describing your standing here is an estimate drawn from samples, which is fine so long as it is labeled.

None of that calls for a separate program with its own budget line. It calls for finishing the ordinary one: facts with dates and figures attached, repeated by sources that are not you, one company name spelled identically everywhere, nothing important behind a form. The same thinking runs through our service pages, and further walkthroughs sit in our blog archive.

To see how this reads on your own domain rather than in the abstract, point the research views at your market: connect a domain and open the market circle. For a company trading here since the Carter administration, the useful output is rarely the score. It is recognizing the version of the business the machine describes, and being able to put a year on it.