Links, authority, and reputation

Internal linking is the cheapest work in this course and the most frequently done badly. Nobody has to approve it, no external party is involved, and the entire surface is under your control. The usual result is a site that adds links by rule rather than by reason.

The rule version looks like this: every article gets three internal links, the anchor text is the target page's keyword, and a plugin inserts them wherever the phrase appears. It satisfies an audit and it produces links that interrupt a reader mid-sentence to offer something they did not want.

A link earns its place when the destination helps the reader continue. More links is not a better site.

Context is what makes a link useful

The same two pages can be linked well or badly depending entirely on where the link sits. Horizon's damages section, where a reader has just learned which losses are recoverable, is exactly where estimate a possible settlement range is a helpful next step. The same link dropped into an unrelated article about choosing a lawyer is an advertisement.

The plugin version is the clearest failure. Inserting best Austin personal injury settlement calculator wherever that phrase occurs produces links nobody clicks, anchors that read like ad copy, and a pattern that looks automated because it is. It also spends the page's credibility on something the reader did not ask for.

Write the anchor to describe what is on the other side, in the words of the surrounding sentence. Estimate a possible settlement range does that. The exact-match keyword version describes the page you wish to rank rather than the page you are linking to, and readers can tell.

Fix the links that already exist before adding more

Broken and redirected internal links are more common than teams expect, and they accumulate silently after every migration. A link to a URL that now 404s wastes the reader's click and the crawl. A link that redirects works, but it is a hop nobody needs and it usually means the source page was never updated after a move.

Fix them at the source, not with a redirect. Adding a redirect to cover an internal link is patching your own site against itself. Update the link to point at the current URL, and keep redirects for external references you do not control.

Do this before any campaign to add new links. A site with forty broken internal links and a plan to add sixty more is prioritizing the visible work over the work that already has an audience.

When no page wants to link to it

The usual way this work gets framed is from the destination: this page needs more internal links, so go and add some. That framing is what produces forced insertions, because it starts from a quota rather than from a reader.

Turn it around. Ask which pages have readers whose next question this page answers. Usually two or three exist and the links write themselves. Sometimes none exist, and that is the finding worth having.

No natural source is an architecture problem, not a linking one. A page nothing on the site has a reason to link to is a page with no place in the map from module 3. It is either serving a task no other page leads to, which means the path into it is missing, or it is serving a task nobody on this site needed, which is a different conversation. Adding forced links buries the question under work.

The same logic caps the count. There is no correct number of internal links per page, and any figure somebody quotes is invented. The limit is how many places in a page a reader's task genuinely changes, which on most pages is a small number and on a long guide might be six.

Classify the anchors instead of summarizing them

Reading a list of anchors and forming an impression produces a sentence like most of them are fine. Counting them into buckets produces a number you can act on, and the two frequently disagree.

Then run the same count on the page's children. This is the comparison that turns a finding into a cheap fix. A parent page whose anchors are 0% descriptive looks like a hard problem. The same count across its seven child pages coming back at 48% descriptive says something else entirely: the site knows how to write a link and does it everywhere except here. That is a habit, and habits are cheaper to change than capabilities.

Improve the paths

Four passes, and the first one identifies the pages worth the effort rather than treating every page equally.

1. Find important pages that are weakly linked.Important against the module 1 outcome. A weakly linked page nobody needs is not a problem, and fixing it is work with no payoff attached.
2. Choose source pages where the reader's task actually changes.Read the candidate section and ask whether somebody there would want the destination next. If the honest answer is no, the link belongs somewhere else or nowhere.
3. Write the anchor as part of the sentence.Descriptive, natural, and specific to the destination. An anchor that reads like it was inserted was inserted, and it is the pattern people notice first.
4. Repair broken and redirected internal links.At the source. Then recheck the crawl paths after release, because this is the kind of change that quietly breaks other things.

Every backlink tool sells a number, and the number is the least useful thing it produces. No commercial index sees every link, they crawl on different schedules, and they disagree with each other by wide margins on the same site. What you are holding is a sample.

Used as a sample it answers real questions. Which relevant sites reference this topic, what happens to the pages they point at, what a competitor earned that we did not, and whether anything here creates a risk worth acting on.

Used as a score it produces the most common bad meeting in this discipline: a number went down, nobody can say which links changed, and a plan gets made to raise the number.

A backlink index is an incomplete sample. Review referring pages and destinations, and never report a vendor score as an engine's own metric.

Read the profile at page level

Not everything in a profile deserves a response. A random low-quality directory is usually noise, and disavowing it is work with no measurable upside. Act when there is a verified risk or an incorrect business claim attached, and otherwise leave it alone.

Name the provider and the scope, every time

Two tools will give you different totals for the same domain on the same day, and both are correct about what they saw. A finding that does not name its provider and its scope cannot be compared against anything later, including a rerun of itself.

The same applies to competitor comparisons. Compare through one provider with identical scope settings, or the difference you are attributing to their link building is partly a difference in what each index happens to have crawled.

Live lookups cost, saved reports do not

Asking the provider what it sees right now spends credits. Reopening a report you already captured does not, and it returns the exact evidence somebody else reviewed rather than a new snapshot that has quietly moved.

Name the report and write the question in its notes. Six weeks later, the difference between a useful record and a mystery is one sentence explaining what you were trying to settle when you ran it. Refresh only when the age of the evidence could genuinely change the decision in front of you.

Work a profile

Six steps. The first one prevents the most common wasted spend, which is re-running an analysis somebody already has.

1. Check saved reports before running anything.Somebody may have captured this last month. Reopening costs nothing and returns the exact evidence, which is better for comparison than a fresh snapshot anyway.
2. Confirm the target scope and name it.Domain, subdomain handling, and provider. Recorded on the report, because a comparison against an unnamed scope is not a comparison.
3. Review referring domains, destinations, and anchors.Read the actual referring pages for the relationships that matter. Aggregates tell you the shape; the pages tell you what to do.
4. Find broken destinations first.Recovering an earned link that now 404s is the cheapest win available in this lesson, and it needs no outreach at all.
5. Inspect new, lost, and anything that looks purchased.Patterns rather than individual rows. One odd link is noise; a hundred sharing an anchor and a footprint is a finding.
6. Compare competitors under identical scope.Same provider, same settings. Look for the asset that earned the links rather than the count, because the asset is the part you can act on.

Link building has a bad reputation because most of it is bad. Bulk outreach to whoever has a contact form, paid placements dressed as editorial, and guest posts written for nobody. All of it is detectable, some of it is risky, and none of it produces a relationship worth having twice.

The version that works is slower and unglamorous. Build something genuinely useful to a specific audience, find the people already writing for that audience, and give them a reason their readers benefit from.

Earn references by being useful to somebody else's readers. Relevance is a property of their audience, not of their domain rating.

The asset has to be worth citing

A reference asset is something a writer would reach for while making their own point. Original data, a working tool, a documented process nobody else has published, or an explanation from somebody genuinely qualified. Horizon's version is a Texas settlement-timeline dataset with named attorneys behind the review.

Notice what makes that citable. It contains numbers nobody else has, it is specific to a place, and it carries names that can be checked. A journalist writing about court delays can use it as evidence. A general article about personal injury law, however well written, gives them nothing to point at.

This is also why the asset comes before the outreach list. Building the list first produces a search for anyone who might link to a page that does not deserve a link yet, which is the shape of every campaign that ends in a spreadsheet of unanswered emails.

Relevance is an audience overlap, not a metric

Overlap on the terms you sell against is a warning, not a match. When a candidate ranks for the same money terms you are targeting, they are a competitor. Competitors do not link to competitors, and a prospect list built from ranking overlap fills up with exactly the sites least likely to respond.

Compensation has to be declared

If money, product, or any other consideration changed hands, the link needs the appropriate attribute and the arrangement needs disclosing. This is not a technicality and it is not optional, and the risk of ignoring it lands on the site being linked to.

The same treatment applies to user-generated and syndicated placements. Handling this honestly costs a small amount of visible caution and removes an entire category of problem that is expensive to unwind later.

Run an outreach cycle

Five steps, and the first two take most of the time. The pitch itself is the easy part once they are done.

1. Build or identify the asset first.Something with original evidence, a working utility, or genuine expertise behind it. Outreach for a page that does not deserve a citation is a volume game you will lose.
2. Find people already covering this for the right audience.Recent, specific, comparable readers. Search for the coverage rather than the domain list, because the coverage is what tells you the fit is real.
3. Write to the specific piece.Name the article, say what your asset adds to the point they were making, and keep it short. A pitch that could have gone to fifty people will be read as one that did.
4. Track replies and decisions, including the refusals.A no with a reason is worth recording. It tells you what the asset is missing and stops somebody pitching the same person next quarter.
5. Handle compensation and attributes correctly.Disclose sponsorship, qualify paid and user-generated links, and keep the record. This is the part that protects the site rather than the campaign.

Mention monitoring produces a feed, and a feed is not useful. A mention can be positive, negative, accurate, wrong, influential, or seen by nobody, and treating all of them as one stream means responding to the loudest rather than the most consequential.

The useful unit is a claim: a specific factual statement about the organization that could be checked. What was said, who said it, when it was seen, which of your approved facts it matches or contradicts, and what follows from that.

The unit is a checkable claim, not a mention. A correction is closed by a retest, not by finishing the task.

Approved facts come first

Comparing a third-party statement against your own record only works if the record is right. Maintain the approved facts with effective dates, and confirm them before trusting any automated verdict, because a stale internal record will confidently flag correct external information as wrong.

Effective dates matter more than people expect. An address that was accurate until March is not an incorrect claim in a January article. It is an out-of-date claim, and the difference decides whether you ask for a correction or leave it alone.

Four verdicts, decided by a person

An opinion is not a false claim. A review saying the service was slow is not something to correct, and treating it as a factual error produces a response that reads badly to everybody watching. Corrections are for checkable facts. Opinions are a different problem with a different remedy.

Fix what you control first

Owned sources come first, always. An outdated address on your own site and business profile undermines every request you are about to send to a directory, and it is the one place you can guarantee the fix.

External corrections are requests, not instructions. Be specific about the fact, provide the evidence, and be straightforward about why it matters to their readers. Most publishers will correct a genuine factual error; almost none respond well to being told what to publish.

The retest is what closes it

Completing a correction task proves your team did its part. It proves nothing about the directory, the publisher, or the answer source, and this is where reputation work most often reports success that has not happened.

Close on a dated retest, not on a completed task. Record what the source says now, where you checked, who checked it, and the date. The outcome is corrected, still present, or unavailable, and all three are legitimate results worth keeping.

Still present after a reasonable interval is information rather than failure. It tells you the request did not land, which is the moment to decide whether to follow up, escalate, or accept it and move on. A task closed without a retest tells you nothing at all, and it quietly enters the record as a success.

Work a claim end to end

Seven steps. The last one is the one that gets dropped, and dropping it is what makes the whole record unreliable.

1. Confirm the approved facts and their effective dates.Before trusting any verdict. A stale internal record produces confident false positives, and those cost credibility with the sources you are about to contact.
2. Collect claims from sources that matter.Saved answers, review sites, forums, video, and web. Filter for consequential rather than comprehensive; a complete feed is a way to do nothing thoroughly.
3. Have a person set the verdict and write the reasoning.Accurate, out of date, incorrect, or not in your record. The reasoning is what makes the decision reviewable when somebody disagrees later.
4. Separate opinions from factual errors.Only the second category is correctable. Sending a correction request about somebody's opinion damages the relationship and the record.
5. Assign a correction owner and a status.With linked work where execution needs accountability. An unowned correction is a note, and notes do not get sent.
6. Correct owned sources, then request external ones.Your own properties first, because they are guaranteed and because they undermine the request if left wrong.
7. Retest and record the outcome.Checked URL, current wording, reviewer, and date. Only a successful retest marks it verified, and anything else is an open item with information attached.